Shenrui-Ma/video-regen-recipes

[Credential-related source omitted]

[Credential-related source omitted]

[Credential-related source omitted]

[Credential-related source omitted]

[Credential-related source omitted]

[Credential-related source omitted]

| 步骤 | 做法 | 关键点 | | --- | --- | --- | | 1. 打开会话 | `opencli browser <session> open "https://www.doubao.com/chat/?cxtarget=<唯一标记>"` | 唯一标记参数让 `osascript` 能精确定位到该标签，避免命中用户自己的其他豆包标签 |

| 步骤 | 做法 | 关键点 | | --- | --- | --- | | 1. 打开会话 | `opencli browser <session> open "https://www.doubao.com/chat/?cxtarget=<唯一标记>"` | 唯一标记参数让 `osascript` 能精确定位到该标签，避免命中用户自己的其他豆包标签 |

| 步骤 | 做法 | 关键点 | | --- | --- | --- | | 1. 打开会话 | `opencli browser <session> open "https://www.doubao.com/chat/?cxtarget=<唯一标记>"` | 唯一标记参数让 `osascript` 能精确定位到该标签，避免命中用户自己的其他豆包标签 |

| --- | --- | --- | | 1. 打开会话 | `opencli browser <session> open "https://www.doubao.com/chat/?cxtarget=<唯一标记>"` | 唯一标记参数让 `osascript` 能精确定位到该标签，避免命中用户自己的其他豆包标签 | | 2. 放图 | 图片进剪贴板（`osascript -e 'set the clipboard to (read (POSIX file "…") as

| --- | --- | --- | | 1. 打开会话 | `opencli browser <session> open "https://www.doubao.com/chat/?cxtarget=<唯一标记>"` | 唯一标记参数让 `osascript` 能精确定位到该标签，避免命中用户自己的其他豆包标签 | | 2. 放图 | 图片进剪贴板（`osascript -e 'set the clipboard to (read (POSIX file "…") as

| --- | --- | --- | | 1. 打开会话 | `opencli browser <session> open "https://www.doubao.com/chat/?cxtarget=<唯一标记>"` | 唯一标记参数让 `osascript` 能精确定位到该标签，避免命中用户自己的其他豆包标签 | | 2. 放图 | 图片进剪贴板（`osascript -e 'set the clipboard to (read (POSIX file "…") as

| 1. 打开会话 | `opencli browser <session> open "https://www.doubao.com/chat/?cxtarget=<唯一标记>"` | 唯一标记参数让 `osascript` 能精确定位到该标签，避免命中用户自己的其他豆包标签 |

| 1. 打开会话 | `opencli browser <session> open "https://www.doubao.com/chat/?cxtarget=<唯一标记>"` | 唯一标记参数让 `osascript` 能精确定位到该标签，避免命中用户自己的其他豆包标签 |

| 1. 打开会话 | `opencli browser <session> open "https://www.doubao.com/chat/?cxtarget=<唯一标记>"` | 唯一标记参数让 `osascript` 能精确定位到该标签，避免命中用户自己的其他豆包标签 |

opencli browser <session> eval "(()=>[...document.querySelectorAll('img')].filter(i=>i.src.includes('byteimg')&&i.naturalWidth>=1000).map(i=>i.src).join('\n'))()"

opencli browser <session> eval "(()=>[...document.querySelectorAll('img')].filter(i=>i.src.includes('byteimg')&&i.naturalWidth>=1000).map(i=>i.src).join('\n'))()"

opencli browser <session> eval "(()=>[...document.querySelectorAll('img')].filter(i=>i.src.includes('byteimg')&&i.naturalWidth>=1000).map(i=>i.src).join('\n'))()"

# 用 python/curl 带 Referer 取回（示例见仓库脚本习惯；不要把链接写进模板）

# 用 python/curl 带 Referer 取回（示例见仓库脚本习惯；不要把链接写进模板）

# 用 python/curl 带 Referer 取回（示例见仓库脚本习惯；不要把链接写进模板）

#   Request(url, headers={'User-Agent':'Mozilla/5.0','Referer':'https://www.doubao.com/'})

#   Request(url, headers={'User-Agent':'Mozilla/5.0','Referer':'https://www.doubao.com/'})

#   Request(url, headers={'User-Agent':'Mozilla/5.0','Referer':'https://www.doubao.com/'})

#!/usr/bin/env python3

#!/usr/bin/env python3

Executable file; content unavailable

Executable file; content unavailable

Executable file; content unavailable

Executable file; content unavailable

Executable file; content unavailable

Executable file; content unavailable

Executable file; content unavailable

Executable file; content unavailable

Executable file; content unavailable

Executable file; content unavailable

Executable file; content unavailable

Executable file

以下是自托管 ComfyUI API，路径附加在运行时 `COMFY_BASE_URL` 后。[官方端点](https://docs.comfy.org/development/comfyui-server/comms_routes)

以 `prompt_id` 隔离任务。WebSocket 的某个节点完成、进度 100%、HTTP 200 都不等于整图成功。轮询采用有限超时与退避，不高频请求。缓存命中是有效结果；需要新候选才改 seed，不为逃避缓存乱改图。[官方 API 示例](https://github.com/Comfy-Org/ComfyUI/blob/master/script_examples/websockets_api_example.py)

SDXL Base 可单独生成，Refiner 是另一个可选阶段。微调模型是否内含兼容 VAE、是否需要特定编码方式，应看该发布版说明。[SDXL 官方模型卡](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0)

Anima 官方示例分别加载扩散模型、Qwen-3 0.6B 文本编码器、Qwen-Image VAE。它们的 loader 参数和 latent 类型以对应版本的官方图与实时 `/object_info` 为准；不要凭节点名自行填 `type`。模板可能使用子图，需展开检查。[ComfyUI Anima 示例](https://docs.comfy.org/tutorials/image/anima/anima)

Anima Base、Aesthetic、Turbo 是不同配置：Base/Aesthetic 的常规采样设置不能移植给蒸馏 Turbo；Aesthetic 的提示词也不应照搬 Base 的整组质量标签。选择精确版本后读取采样与提示词说明，记录实际值，不自动追新。[发布方模型卡](https://huggingface.co/circlestone-labs/Anima)

- **SDXL：**先读具体 checkpoint 的示例和标签习惯。用少量准确标签或短句；仅保留该模型需要的质量标签与 LoRA 触发词。不把某个微调分支的评分标签推广到全部 SDXL。 - **Anima：**可使用标签、完整自然语言或混合提示。使用自然语言时把外观与动作关系写清楚；按所选版本决定质量标签。不要照搬 SDXL 权重强度或某个版本的整段负面词。[Anima 提示说明](https://huggingface.co/circlestone-labs/Anim

- **Anima：**可使用标签、完整自然语言或混合提示。使用自然语言时把外观与动作关系写清楚；按所选版本决定质量标签。不要照搬 SDXL 权重强度或某个版本的整段负面词。[Anima 提示说明](https://huggingface.co/circlestone-labs/Anima#prompting)

1. 使用已有 ComfyUI；从用户配置或正在使用的实例确定地址。未指定时先探测 `http://127.0.0.1:8188`，不要假定桌面版使用同一端口。

2. 请求 `GET /system_stats`、`GET /object_info`、`GET /queue`。确认返回 JSON，而非登录页或 HTML 错误；只记录必要设备能力，不公开原始系统信息。 3. 不可达时，检查用户指定的安装方式、端口和启动状态。若需要安装，先按[官方本地说明](https://docs.comfy.org/installation/system_requirements)确认设备支持及依赖，不盲装节点包或覆盖现有环境。 4. 定义运行时变量

3. 不可达时，检查用户指定的安装方式、端口和启动状态。若需要安装，先按[官方本地说明](https://docs.comfy.org/installation/system_requirements)确认设备支持及依赖，不盲装节点包或覆盖现有环境。

**Civitai（C 站）：**使用官方 `GET https://civitai.com/api/v1/models`，通过 `query`、`types`、`baseModels` 缩小候选。枚举值按当前文档读取，关键词要 URL 编码；带 `query` 时用返回的 cursor 分页，不能同时带 `page`。搜索结果只作候选。[模型搜索 API](https://github.com/civitai/civitai-developer-docs/blob/main

读取 `GET /api/v1/models/{model_id}` 的作者说明及许可，再读 `GET /api/v1/model-versions/{version_id}`：核对 `baseModel`、版本说明、触发词、全部 `files` 的格式/精度/大小/完整 SHA-256，选定具体文件及其 `downloadUrl`。不要默认首个文件、最高下载量或模型级链接指向所需版本。已有 SHA-256 可用 `/api/v1/model-versions/by-hash

**Hugging Face：**优先模型发布方仓库，再看其链接的兼容发布。可用站内搜索，或现有 `huggingface_hub` 的 `HfApi().list_models(search=..., limit=...)`。读模型卡、配置、文件列表及 LICENSE；用 `model_info(repo_id, files_metadata=True)` 取得版本信息，将分支解析成 commit SHA，再按该 commit 核查具体文件。仓库名和标签不能单独证明架构兼容

1. 确认版本许可覆盖用户的用法；模型权重许可和生成图片许可分开记录。下载权限不等于转载许可。无需将模型权重提交本仓库。 2. 校验目标目录、预计大小和空间，包括暂存与搬运占用。选择可信来源的安全张量文件；不要加载不可信 pickle，也不要为了读模型元数据执行仓库代码。 3. HF 优先使用已有官方客户端的单文件下载，指定 `repo_id`、`filename` 和固定 `revision`；支持时先 `dry_run=True`。恢复时沿用相同缓存/目录及 revisi

2. 校验目标目录、预计大小和空间，包括暂存与搬运占用。选择可信来源的安全张量文件；不要加载不可信 pickle，也不要为了读模型元数据执行仓库代码。 3. HF 优先使用已有官方客户端的单文件下载，指定 `repo_id`、`filename` 和固定 `revision`；支持时先 `dry_run=True`。恢复时沿用相同缓存/目录及 revision，不下载整库。[官方文件下载](https://huggingface.co/docs/huggingface_hub

3. HF 优先使用已有官方客户端的单文件下载，指定 `repo_id`、`filename` 和固定 `revision`；支持时先 `dry_run=True`。恢复时沿用相同缓存/目录及 revision，不下载整库。[官方文件下载](https://huggingface.co/docs/huggingface_hub/guides/download)

{   "schema_version": 1,   "repository": "https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit",

"schema_version": 1,   "repository": "https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit",   "workflows": [

"repository": "https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit",

涉及 ComfyUI 时，先检查 [shenrui-comfyui-toolkit](https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit) 的[工作流目录](https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit/blob/main/workflows/index.json)及对应说明。该仓库维护具体节点图和执行工具；Recipes 负责题材、提示词、输入资产、时间线和

#!/usr/bin/env python3

#!/usr/bin/env python3

else:         if offline:raise ValueError('Offline mode needs a verified cache file or a toolkit checkout.')         if lock['repository']!='https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit' or not re.fullmatch('[0-9a-f]{40}',lock['rev

if offline:raise ValueError('Offline mode needs a verified cache file or a toolkit checkout.')         if lock['repository']!='https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit' or not re.fullmatch('[0-9a-f]{40}',lock['revision']):     

if lock['repository']!='https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit' or not re.fullmatch('[0-9a-f]{40}',lock['revision']):

raise ValueError('Unexpected toolkit repository or revision.')         url='https://raw.githubusercontent.com/Shenrui-Ma/shenrui-comfyui-toolkit/'+lock['revision']+'/'+relative.as_posix()         with urllib.request.urlopen(url,timeout=30) 

url='https://raw.githubusercontent.com/Shenrui-Ma/shenrui-comfyui-toolkit/'+lock['revision']+'/'+relative.as_posix()         with urllib.request.urlopen(url,timeout=30) as response:data=response.read(2*1024*1024+1)     if len(data)>2*1024*1

with urllib.request.urlopen(url,timeout=30) as response:data=response.read(2*1024*1024+1)

[Credential-related source omitted]

Executable file

for field in ("model", "prefix"):             for value in ("", "   ", None, "/absolute/file", "../file", "C:\\file",                           "dir/../file", "dir//file", "https://host/file"):

for value in ("", "   ", None, "/absolute/file", "../file", "C:\\file",                           "dir/../file", "dir//file", "https://host/file"):                 with self.subTest(field=field, value=value), self.assertRaises(ValueError):

"dir/../file", "dir//file", "https://host/file"):

[Credential-related source omitted]

[Credential-related source omitted]

顶层 `model` 是有图像工具能力的语言模型；工具内部的 `model` 才是 `gpt-image-2`。`quality=high` 是图像工具参数，与语言模型的 reasoning effort 无关。使用 `input_image` 附加图片，并要求 `image_generation` 工具调用；输出读取 `image_generation_call.result`，同时保存可用的 `revised_prompt`。[官方图像工具说明](https://deve

[Credential-related source omitted]

[Credential-related source omitted]

本客户端将 `store=false`，不依赖服务端保存后再取回。非流式请求减少误把预览图当成终图的分支；`max_tool_calls=1` 限制内置工具总调用次数，一次只接受一个完成的图像结果。[Responses 参数](https://developers.openai.com/api/reference/cli/resources/responses/methods/create)更复杂的流式实现必须区分 partial 与 completed，不照搬“最后收到的 

锁定 [GPT Image 2](https://developers.openai.com/api/docs/models/gpt-image-2)，不跟随其他模型的默认值。按[官方尺寸规则](https://developers.openai.com/api/docs/guides/image-generation#earlier-gpt-image-models)：两边为 16 的倍数、最大边 ≤3840、长宽比 ≤3、像素总数 655360–8294400。客户端要求

先确认图像用途，再写主体、动作、构图、光线和保留项。使用清楚的自然语言，参考图承担外观证据；避免仅堆角色名、形容词或互相冲突的约束。多图需要指定各自职责。[官方提示词指南](https://developers.openai.com/api/docs/guides/image-prompting)

#!/usr/bin/env python3

ENDPOINT = "https://api.openai.com/v1/responses"

def post_once(payload, key):     request = urllib.request.Request(ENDPOINT, data=payload, headers={

request = urllib.request.Request(ENDPOINT, data=payload, headers={

data = base64.b64decode(call["result"], validate=True)

[Credential-related source omitted]

[Credential-related source omitted]

Executable file

return gen.PNG + chunk(b"IHDR", struct.pack(">IIBBBBB", 1, 1, 8, 2, 0, 0, 0)) + chunk(b"IDAT", zlib.compress(b"\x00\xff\x00\x00")) + chunk(b"IEND", b"")

[Credential-related source omitted]

[Credential-related source omitted]

[Credential-related source omitted]

[Credential-related source omitted]

[Credential-related source omitted]

[Credential-related source omitted]

[Credential-related source omitted]

[Credential-related source omitted]

[Credential-related source omitted]

[Credential-related source omitted]

[Credential-related source omitted]

[Credential-related source omitted]

[Credential-related source omitted]

**模型展示名不等于 API ID。** 当前官方模型页面列出 V5、V4.5、V4 等系列；Image OpenAPI 的 `model` 是普通字符串，没有完整模型枚举，仅在标签接口示例出现 `nai-diffusion-3`。不能据此推导 V5/V4.5 的 ID 或默认退回 V3。[官方模型页面](https://docs.novelai.net/en/image/models/) · [官方 schema](https://image.novelai.net/doc

POST https://image.novelai.net/ai/generate-image

核心参数是宽高、张数、steps、scale、sampler、seed、negative_prompt。结构化提示可使用 `v4_prompt` / `v4_negative_prompt`，包含 `caption.base_caption` 与 `caption.char_captions`；名字中有 v4 不代表可随意套用所有代际。图生图需要 `image` 的图片 base64 和 `strength`、`noise` 等适配参数，不能把本地路径或 data URL 当

客户端采用非流式 ZIP 输出。当前 schema 也提供 JSON 图片响应和 SSE 接口，本客户端不使用它们。`x-correlation-id` 是排查请求用的六位字母数字标识，**不是防重复计费的幂等键**。[官方 API](https://image.novelai.net/docs/index.html)

**Vibe Transfer：** 官方提供 `POST /ai/encode-vibe`，输出二进制编码；生成参数有 `reference_image_multiple`、`reference_information_extracted_multiple`、`reference_strength_multiple`。V4 及以上有编码缓存，改变提取量可能再次计费。不要将整份 `.naiv4vibe` 文件或其 JSON 当作纯编码塞入数组。要核实该模型需要原图还是编码，以

**Precise Reference：** schema 提供 `director_reference_images`、`director_reference_descriptions`、`director_reference_information_extracted`、`director_reference_strength_values` 与 `director_reference_secondary_strength_values`；数组必须逐项对应。API 的 C

图生图仍需描述要保留的可见内容；`strength` 越低通常越接近底图，但不是像素锁定。[官方图生图说明](https://docs.novelai.net/en/image/controltools/)

V4/V4.5 的标签和语言约束与 V5 不完全相同。选定模型后查对应官方说明；不要将某代提示词格式直接复制到另一代。结构化多角色提示中，基础场景和各人物描述分开；使用坐标时核对对应人物。API 请求的平铺提示词和结构化提示词必须同步修改。[官方模型说明](https://docs.novelai.net/en/image/models/) · [多角色提示](https://docs.novelai.net/en/image/multiplecharacters/)

#!/usr/bin/env python3

ENDPOINT = "https://image.novelai.net/ai/generate-image"

raw = base64.b64decode(p.get("image", ""), validate=True)

[Credential-related source omitted]

[Credential-related source omitted]

[Credential-related source omitted]

[Credential-related source omitted]

Executable file

def test_redirect_refused(self):         with self.assertRaises(api.SafeError):             api.NoRedirect().redirect_request(None, None, 302, "", {}, "https://elsewhere.example")

api.NoRedirect().redirect_request(None, None, 302, "", {}, "https://elsewhere.example")

[tls] IO error: End of file Error opening input file https://upos-sz-estgcos.bilivideo.com/...

Error opening input file https://upos-sz-estgcos.bilivideo.com/...

#!/usr/bin/env python3

match = re.fullmatch(r'/video/(BV[A-Za-z0-9]{10}|av[0-9]+)/?', parsed.path)         if not match:             raise FetchError(EXIT_USAGE, '只支持 B 站普通视频页面，如 https://www.bilibili.com/video/BVxxxx/')

if not match:             raise FetchError(EXIT_USAGE, '只支持 B 站普通视频页面，如 https://www.bilibili.com/video/BVxxxx/')         parts = urllib.parse.parse_qs(parsed.query, keep_blank_values=True).get('p', [])

raise FetchError(EXIT_USAGE, '只支持 B 站普通视频页面，如 https://www.bilibili.com/video/BVxxxx/')         parts = urllib.parse.parse_qs(parsed.query, keep_blank_values=True).get('p', [])         if parts and (len(parts) != 1 or not re.fullmatch(r'[1-9

raise FetchError(EXIT_USAGE, 'p 参数必须是单个正整数')         suffix = '?p=' + parts[0] if parts else ''         return 'bilibili', 'https://www.bilibili.com/video/' + match.group(1) + '/' + suffix

suffix = '?p=' + parts[0] if parts else ''         return 'bilibili', 'https://www.bilibili.com/video/' + match.group(1) + '/' + suffix     if host in YOUTUBE_HOSTS:

return 'bilibili', 'https://www.bilibili.com/video/' + match.group(1) + '/' + suffix

ids = urllib.parse.parse_qs(parsed.query).get('v', [])             if len(ids) == 1 and re.fullmatch(r'[A-Za-z0-9_-]{11}', ids[0]):                 return 'youtube', 'https://www.youtube.com/watch?v=' + ids[0]

if len(ids) == 1 and re.fullmatch(r'[A-Za-z0-9_-]{11}', ids[0]):                 return 'youtube', 'https://www.youtube.com/watch?v=' + ids[0]         if host == 'youtu.be':

return 'youtube', 'https://www.youtube.com/watch?v=' + ids[0]

candidate = parsed.path.strip('/')             if re.fullmatch(r'[A-Za-z0-9_-]{11}', candidate):                 return 'youtube', 'https://www.youtube.com/watch?v=' + candidate

if re.fullmatch(r'[A-Za-z0-9_-]{11}', candidate):                 return 'youtube', 'https://www.youtube.com/watch?v=' + candidate         raise FetchError(EXIT_USAGE, '只支持 YouTube 普通 watch 链接')

return 'youtube', 'https://www.youtube.com/watch?v=' + candidate

request = urllib.request.Request(url, headers={'User-Agent': 'Mozilla/5.0 (reference-video-fetch)'})

try:         response = (opener or urllib.request.urlopen)(request, timeout=20)         final = response.geturl() if hasattr(response, 'geturl') else response

response = (opener or urllib.request.urlopen)(request, timeout=20)

def fetch(args, *, runner=subprocess.run, opener=None):

args = parser().parse_args(argv)     try:         code, record = fetch(args)

try:         code, record = fetch(args)     except FetchError as error:

code, record = fetch(args)

Executable file

def args_for(**overrides):     base = {         "url": "https://www.bilibili.com/video/BV1LQ7j6WEyD", "out": None, "stem": None, "record": None,

base = {         "url": "https://www.bilibili.com/video/BV1LQ7j6WEyD", "out": None, "stem": None, "record": None,         "max_height": 1080, "require_height": None, "max_duration": 900, "limit_rate": None,

"url": "https://www.bilibili.com/video/BV1LQ7j6WEyD", "out": None, "stem": None, "record": None,

class UrlTests(unittest.TestCase):     def test_accepts_bilibili_and_youtube_pages(self):         self.assertEqual(fetch_reference.canonical_page("https://www.bilibili.com/video/BV1LQ7j6WEyD/"),

def test_accepts_bilibili_and_youtube_pages(self):         self.assertEqual(fetch_reference.canonical_page("https://www.bilibili.com/video/BV1LQ7j6WEyD/"),                          ("bilibili", "https://www.bilibili.com/video/BV1LQ7j6WEyD/"

self.assertEqual(fetch_reference.canonical_page("https://www.bilibili.com/video/BV1LQ7j6WEyD/"),

("bilibili", "https://www.bilibili.com/video/BV1LQ7j6WEyD/"))

self.assertEqual(fetch_reference.canonical_page("https://m.bilibili.com/video/av12345?p=2"),

("bilibili", "https://www.bilibili.com/video/av12345/?p=2"))

self.assertEqual(             fetch_reference.canonical_page("https://www.bilibili.com/video/BV1LQ7j6WEyD/?spm_id_from=333.999"),             ("bilibili", "https://www.bilibili.com/video/BV1LQ7j6WEyD/"),

fetch_reference.canonical_page("https://www.bilibili.com/video/BV1LQ7j6WEyD/?spm_id_from=333.999"),

("bilibili", "https://www.bilibili.com/video/BV1LQ7j6WEyD/"),

)         self.assertEqual(fetch_reference.canonical_page("https://www.youtube.com/watch?v=dQw4w9WgXcQ"),                          ("youtube", "https://www.youtube.com/watch?v=dQw4w9WgXcQ"))

self.assertEqual(fetch_reference.canonical_page("https://www.youtube.com/watch?v=dQw4w9WgXcQ"),

("youtube", "https://www.youtube.com/watch?v=dQw4w9WgXcQ"))

self.assertEqual(fetch_reference.canonical_page("https://youtu.be/dQw4w9WgXcQ"),

("youtube", "https://www.youtube.com/watch?v=dQw4w9WgXcQ"))

def test_resolves_short_links_through_the_opener(self):         opener = lambda request, timeout=None: type("R", (), {             "geturl": lambda self: "https://www.bilibili.com/video/BV1LQ7j6WEyD/",

opener = lambda request, timeout=None: type("R", (), {             "geturl": lambda self: "https://www.bilibili.com/video/BV1LQ7j6WEyD/",             "close": lambda self: None,

"geturl": lambda self: "https://www.bilibili.com/video/BV1LQ7j6WEyD/",

})()         self.assertEqual(             fetch_reference.canonical_page("https://b23.tv/abcDEF", opener=opener),

self.assertEqual(             fetch_reference.canonical_page("https://b23.tv/abcDEF", opener=opener),             ("bilibili", "https://www.bilibili.com/video/BV1LQ7j6WEyD/"),

fetch_reference.canonical_page("https://b23.tv/abcDEF", opener=opener),

("bilibili", "https://www.bilibili.com/video/BV1LQ7j6WEyD/"),

def test_rejects_media_links_and_other_hosts(self):         for url in (             "https://upos-sz-estgcos.bilivideo.com/upgcxcode/1/2.m4s?e=abc",

for url in (             "https://upos-sz-estgcos.bilivideo.com/upgcxcode/1/2.m4s?e=abc",             "https://www.bilibili.com/bangumi/play/ep123456",

"https://upos-sz-estgcos.bilivideo.com/upgcxcode/1/2.m4s?e=abc",

"https://www.bilibili.com/bangumi/play/ep123456",

"ftp://www.bilibili.com/video/BV1LQ7j6WEyD/",             "https://example.com/video/BV1LQ7j6WEyD/",         ):

"https://example.com/video/BV1LQ7j6WEyD/",

def test_rejects_credentials_in_url(self):         with self.assertRaises(fetch_reference.FetchError):             fetch_reference.canonical_page("https://user:pass@www.bilibili.com/video/BV1LQ7j6WEyD/")

fetch_reference.canonical_page("https://user:pass@www.bilibili.com/video/BV1LQ7j6WEyD/")

chosen, audio, _ = fetch_reference.choose_formats(PORTRAIT_INFO, 1080)         command = fetch_reference.build_command(             "/fake/yt-dlp", "https://www.bilibili.com/video/BV1LQ7j6WEyD/", chosen, audio,

command = fetch_reference.build_command(             "/fake/yt-dlp", "https://www.bilibili.com/video/BV1LQ7j6WEyD/", chosen, audio,             Path("/tmp/out/BV1LQ7j6WEyD.%(ext)s"), timeout=120, limit_rate="4M", sleep_requests=1,

"/fake/yt-dlp", "https://www.bilibili.com/video/BV1LQ7j6WEyD/", chosen, audio,

self.assertNotIn("--cookies", command)         self.assertNotIn("--cookies-from-browser", command)         self.assertEqual(command[-1], "https://www.bilibili.com/video/BV1LQ7j6WEyD/")

[Credential-related source omitted]

self.assertEqual(command[-1], "https://www.bilibili.com/video/BV1LQ7j6WEyD/")

with tempfile.TemporaryDirectory() as temp:             record_path = Path(temp) / "probe.json"             code, record = fetch_reference.fetch(

record_path = Path(temp) / "probe.json"             code, record = fetch_reference.fetch(                 args_for(out=temp, probe=True, record=record_path), runner=runner

code, record = fetch_reference.fetch(

runner = FakeRunner()         with tempfile.TemporaryDirectory() as temp:             code, record = fetch_reference.fetch(args_for(out=temp), runner=runner)

with tempfile.TemporaryDirectory() as temp:             code, record = fetch_reference.fetch(args_for(out=temp), runner=runner)             self.assertEqual(code, fetch_reference.EXIT_OK)

code, record = fetch_reference.fetch(args_for(out=temp), runner=runner)

(Path(temp) / "BV1LQ7j6WEyD.mp4").write_bytes(b"already-here")             with self.assertRaises(fetch_reference.FetchError) as caught:                 fetch_reference.fetch(args_for(out=temp), runner=runner)

with self.assertRaises(fetch_reference.FetchError) as caught:                 fetch_reference.fetch(args_for(out=temp), runner=runner)             self.assertEqual(caught.exception.code, fetch_reference.EXIT_EXISTS)

fetch_reference.fetch(args_for(out=temp), runner=runner)

with tempfile.TemporaryDirectory() as temp:             with self.assertRaises(fetch_reference.FetchError) as caught:                 fetch_reference.fetch(args_for(out=temp, require_height=2160), runner=runner)

with self.assertRaises(fetch_reference.FetchError) as caught:                 fetch_reference.fetch(args_for(out=temp, require_height=2160), runner=runner)             self.assertEqual(caught.exception.code, fetch_reference.EXIT_QUALITY)

fetch_reference.fetch(args_for(out=temp, require_height=2160), runner=runner)

runner = FakeRunner(info=info)         with tempfile.TemporaryDirectory() as temp:             _, record = fetch_reference.fetch(args_for(out=temp), runner=runner)

with tempfile.TemporaryDirectory() as temp:             _, record = fetch_reference.fetch(args_for(out=temp), runner=runner)             self.assertTrue(record["quality_limited"])

_, record = fetch_reference.fetch(args_for(out=temp), runner=runner)

with patch.object(fetch_reference.shutil, "which", lambda name: None):             with self.assertRaises(fetch_reference.FetchError) as caught:                 fetch_reference.fetch(args_for(yt_dlp=None), runner=FakeRunner())

with self.assertRaises(fetch_reference.FetchError) as caught:                 fetch_reference.fetch(args_for(yt_dlp=None), runner=FakeRunner())         self.assertEqual(caught.exception.code, fetch_reference.EXIT_NO_TOOL)

fetch_reference.fetch(args_for(yt_dlp=None), runner=FakeRunner())

with tempfile.TemporaryDirectory() as temp:             with self.assertRaises(fetch_reference.FetchError) as caught:                 fetch_reference.fetch(args_for(out=temp), runner=runner)

with self.assertRaises(fetch_reference.FetchError) as caught:                 fetch_reference.fetch(args_for(out=temp), runner=runner)             self.assertEqual(caught.exception.code, fetch_reference.EXIT_TOO_LONG)

fetch_reference.fetch(args_for(out=temp), runner=runner)

with tempfile.TemporaryDirectory() as temp:             with self.assertRaises(fetch_reference.FetchError) as caught:                 fetch_reference.fetch(args_for(out=temp), runner=runner)

with self.assertRaises(fetch_reference.FetchError) as caught:                 fetch_reference.fetch(args_for(out=temp), runner=runner)             self.assertEqual(caught.exception.code, fetch_reference.EXIT_UNAVAILABLE)

fetch_reference.fetch(args_for(out=temp), runner=runner)

runner = FakeRunner(info=info)         with tempfile.TemporaryDirectory() as temp:             _, record = fetch_reference.fetch(args_for(out=temp), runner=runner)

with tempfile.TemporaryDirectory() as temp:             _, record = fetch_reference.fetch(args_for(out=temp), runner=runner)             self.assertEqual(record["site"], "bilibili")

_, record = fetch_reference.fetch(args_for(out=temp), runner=runner)

with tempfile.TemporaryDirectory() as temp:             with self.assertRaises(fetch_reference.FetchError) as caught:                 fetch_reference.fetch(args_for(out=temp), runner=runner)

with self.assertRaises(fetch_reference.FetchError) as caught:                 fetch_reference.fetch(args_for(out=temp), runner=runner)             self.assertEqual(caught.exception.code, fetch_reference.EXIT_FAILED)

fetch_reference.fetch(args_for(out=temp), runner=runner)

def test_record_keeps_the_documented_keys(self):         record = fetch_reference.build_record(             PORTRAIT_INFO, "bilibili", "https://www.bilibili.com/video/BV1LQ7j6WEyD/", notes=[],

record = fetch_reference.build_record(             PORTRAIT_INFO, "bilibili", "https://www.bilibili.com/video/BV1LQ7j6WEyD/", notes=[],             path=None, selected=PORTRAIT_INFO["formats"][2], audio=PORTRAIT_INFO["formats"][4],

PORTRAIT_INFO, "bilibili", "https://www.bilibili.com/video/BV1LQ7j6WEyD/", notes=[],

def test_uncapped_record_says_so_instead_of_pretending_a_cap(self):         record = fetch_reference.build_record(             PORTRAIT_INFO, "bilibili", "https://www.bilibili.com/video/BV1LQ7j6WEyD/", notes=[],

record = fetch_reference.build_record(             PORTRAIT_INFO, "bilibili", "https://www.bilibili.com/video/BV1LQ7j6WEyD/", notes=[],             path=None, selected=PORTRAIT_INFO["formats"][2], audio=None,

PORTRAIT_INFO, "bilibili", "https://www.bilibili.com/video/BV1LQ7j6WEyD/", notes=[],

示例：[AI 少女乐队番剧（B站）](https://www.bilibili.com/video/BV1WhY364ECX)。提示词无任何指导，单次抽卡，有瑕疵。

- [shenrui-comfyui-toolkit](https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit)：`h3/ref2va-dialogue` 工作流，固定 revision 与文件哈希见 [workflows/toolkit-source.json](workflows/toolkit-source.json)

- [Shenrui-Ma/video-regen-assets](https://huggingface.co/datasets/Shenrui-Ma/video-regen-assets)：公开素材库；角色图与音色按用户自己的来源准备

从 [MyGO 官方人物目录](https://anime.bang-dream.com/mygo/character/)或 [Ave Mujica 官方人物目录](https://anime.bang-dream.com/avemujica/character/)进入具体人物页，核对日文名、外观、乐器和当前服装。搜索结果的缩略图不能代替身份核验。

1. **找来源。** 从动画官网及其链接的官方频道找人物介绍或 PV；用日文名与 `キャラ別紹介CM` 辅助检索。核对频道、标题和说话角色，不能盲取搜索第一条。取得可用音轨后记录媒体 URL、格式与哈希；音频容器不应标成完整视频。 2. **转写完整音轨。** 原流程使用 faster-whisper `small`、CPU `int8`，日语 `language="ja"`、`beam_size=5`、`word_timestamps=True`、`vad_filter=

2. **转写完整音轨。** 原流程使用 faster-whisper `small`、CPU `int8`，日语 `language="ja"`、`beam_size=5`、`word_timestamps=True`、`vad_filter=False`。保存句子和词时间戳。需实际遍历返回的 segments 才会完成转写；模型缺失时先准备模型，不能同时要求离线且假设已有缓存。[faster-whisper 用法](https://github.com/SYSTRAN/f

原流程是 `pymss==2.0.14`、`pymss-core==0.1.4`，Bandit+ 模型 `model_bandit_plus_dnr_sdr_11.47`，配套同名 checkpoint 与 YAML。当前入口以[该版本发布文档](https://pypi.org/project/pymss/2.0.14/)及实际安装版本的帮助为准；先在独立环境核对模型目录，先处理一人，再批量其余角色。

`setpts` 在这里重建约定的逐帧时间戳；它不是补帧或插帧。输入若本来是可变帧率，先处理时间线合同，不能用这个示例静默改变节奏。[FFmpeg 滤镜说明](https://ffmpeg.org/ffmpeg-filters.html#setpts_002c-asetpts)

`ass` 烧录依赖启用 libass 的 FFmpeg；路径含滤镜特殊字符时由 Agent 正确转义，或将后期文件放入本项目的简单相对路径。[FFmpeg ASS 文档](https://ffmpeg.org/ffmpeg-filters.html#ass)

具体图由 [ComfyUI Toolkit](https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit/tree/main/workflows/h3/ref2va-dialogue)维护；本包保留[固定来源](../workflows/toolkit-source.json)对应的离线副本，便于构建器稳定使用。更新时先核对commit与哈希，不自动漂移。

运行器使用同一套 ComfyUI API：默认连接本机 `http://127.0.0.1:8188`；远程使用者在当前终端设置 `COMFY_BASE_URL` 为自己的服务地址。服务、模型和输入素材都应已准备好，电脑上需有 Python 3、`ffprobe`、`ffmpeg`。支持 macOS、Linux、Windows。

[Credential-related source omitted]

接口依据：[ComfyUI 服务路由](https://github.com/comfyanonymous/ComfyUI/blob/master/server.py)、[SaveVideo](https://github.com/comfyanonymous/ComfyUI/blob/master/comfy_extras/nodes_video.py)、[PreviewVideo 输出结构](https://github.com/comfyanonymous/ComfyU

#!/usr/bin/env python3

#!/usr/bin/env python3

#!/usr/bin/env python3

from contextlib import contextmanager import hashlib import http.client

import hashlib import http.client import json

import http.client

if data is not None:             outgoing["Content-Type"] = "application/json"         request = urllib.request.Request(self.base_url + path, data=data, headers=outgoing)

outgoing["Content-Type"] = "application/json"         request = urllib.request.Request(self.base_url + path, data=data, headers=outgoing)         try:

request = urllib.request.Request(self.base_url + path, data=data, headers=outgoing)

except urllib.error.HTTPError as error:             raise ShotError(f"ComfyUI returned HTTP {error.code}; inspect its UI without logging credentials.") from None         except (urllib.error.URLError, OSError, http.client.HTTPException):

raise ShotError(f"ComfyUI returned HTTP {error.code}; inspect its UI without logging credentials.") from None         except (urllib.error.URLError, OSError, http.client.HTTPException):             raise Pending("Connection interrupted; res

except (urllib.error.URLError, OSError, http.client.HTTPException):

raise ShotError("API JSON exceeds the 64 MiB safety limit.")                 result = json.loads(raw)             except (ValueError, http.client.HTTPException, OSError):

result = json.loads(raw)             except (ValueError, http.client.HTTPException, OSError):                 raise Pending("Incomplete or invalid JSON response; resume the existing run.") from None

except (ValueError, http.client.HTTPException, OSError):

raise Pending("Download is incomplete; resume to fetch the same output again.")         temporary.replace(destination)     except (OSError, http.client.HTTPException):

temporary.replace(destination)     except (OSError, http.client.HTTPException):         raise Pending("Output transfer interrupted; resume to fetch the same output again.") from None

except (OSError, http.client.HTTPException):         raise Pending("Output transfer interrupted; resume to fetch the same output again.") from None     finally:

if not all(shutil.which(tool) for tool in ("ffprobe", "ffmpeg")):             raise ShotError("Install ffprobe and ffmpeg and verify they are on PATH before running.")         client = Client(os.environ.get("COMFY_BASE_URL", "http://127.0.0

raise ShotError("Install ffprobe and ffmpeg and verify they are on PATH before running.")         client = Client(os.environ.get("COMFY_BASE_URL", "http://127.0.0.1:8188"),                         args.wait_seconds, os.environ.get("COMFY_AU

[Credential-related source omitted]

[Credential-related source omitted]

[Credential-related source omitted]

- [BanG Dream! It's MyGO!!!!! 官方人物目录](https://anime.bang-dream.com/mygo/character/)

- [BanG Dream! Ave Mujica 官方人物目录](https://anime.bang-dream.com/avemujica/character/)

- [Ave Mujica 官方角色与演员名单](https://anime.bang-dream.com/avemujica/staff-cast/)

- [MiniMax H3 模型卡](https://huggingface.co/MiniMaxAI/MiniMax-H3)及[官方提示词规范](https://github.com/MiniMax-AI/MiniMax-H3/tree/main/skills/h3-prompt-writing)

- [ComfyUI API](https://docs.comfy.org/development/comfyui-server/comms_routes)

- [faster-whisper](https://github.com/SYSTRAN/faster-whisper)与[WhisperX](https://github.com/m-bain/whisperX)

- [pymss 2.0.14 发布说明](https://pypi.org/project/pymss/2.0.14/)与[历史分离模型来源](https://huggingface.co/baicai1145/pymss)

- [FFmpeg 滤镜文档](https://ffmpeg.org/ffmpeg-filters.html)

- 原制作记录注明曾参考[分镜、音频、风格与提示词教程](https://www.bilibili.com/video/BV1gTbW6oEUh/)。本次未重新观看，不把其建议当作本包已运行的节点。

Executable file

Executable file

def test_binding_escapes_urls_and_annotations_fail(self):         entry = next(iter(self.bindings["files"].values()))         for name in ("/tmp/file.wav", "../file.wav", "C:\\input\\file.wav", "https://example.org/file.wav", "x.wav [output

entry = next(iter(self.bindings["files"].values()))         for name in ("/tmp/file.wav", "../file.wav", "C:\\input\\file.wav", "https://example.org/file.wav", "x.wav [output]"):             with self.subTest(name=name):

for name in ("/tmp/file.wav", "../file.wav", "C:\\input\\file.wav", "https://example.org/file.wav", "x.wav [output]"):

Executable file

class TransportAndMediaTests(unittest.TestCase):     def test_url_rejects_embedded_secrets(self):         for url in ("https://user:secret@example.test", "https://example.test?key=secret", "file:///private"):

def test_url_rejects_embedded_secrets(self):         for url in ("https://user:secret@example.test", "https://example.test?key=secret", "file:///private"):             with self.assertRaises(runner.ShotError) as caught:

for url in ("https://user:secret@example.test", "https://example.test?key=secret", "file:///private"):

[Credential-related source omitted]

[Credential-related source omitted]

def test_redirect_is_refused(self):         with self.assertRaisesRegex(runner.ShotError, "redirect refused"):             runner.NoRedirect().redirect_request(None, None, 302, "", {}, "https://other.test")

runner.NoRedirect().redirect_request(None, None, 302, "", {}, "https://other.test")

[Credential-related source omitted]

[Credential-related source omitted]

client.json("/queue")         request = opened.call_args.args[0]         self.assertEqual(request.full_url, "https://example.test:443/comfy/queue")

request = opened.call_args.args[0]         self.assertEqual(request.full_url, "https://example.test:443/comfy/queue")         self.assertEqual(request.get_header("Authorization"), "Bearer secret")

self.assertEqual(request.full_url, "https://example.test:443/comfy/queue")

{   "repository": "https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit",

"repository": "https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit",

- ComfyUI Core、公开MotionContext、Sage及相关runtime组件：具体版本、来源和许可证副本见 [环境依赖锁](https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit/tree/667eafddb9f42bb6c72ad27b665def9ada36df45/environments/h3/dependencies.lock.json) 与 Toolkit 的 `environments/h3/

- [shenrui-comfyui-toolkit](https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit)

- [Shenrui-Ma/video-regen-assets](https://huggingface.co/datasets/Shenrui-Ma/video-regen-assets)

全部命令从本目录运行。默认用用户已有的 ComfyUI 和模型；只有缺依赖或推理报错需要排查版本时，才查[环境参考记录](https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit/tree/667eafddb9f42bb6c72ad27b665def9ada36df45/environments/h3/README.md)，按需取用其中一段。确实要另建一套时用不存在的目标目录，绝不覆盖共享环境。模型已齐全且可用时直接复用，不重

python3 <toolkit>/environments/h3/scripts/preflight.py    --comfy-root ./h3-local/ComfyUI    --python-site ./h3-local/.venv/lib/python3.10/site-packages    --url http://127.0.0.1:8188 --verify-model-hashes

--comfy-root ./h3-local/ComfyUI \   --python-site ./h3-local/.venv/lib/python3.10/site-packages \   --url http://127.0.0.1:8188 --verify-model-hashes

--url http://127.0.0.1:8188 --verify-model-hashes

python3 scripts/runtime/heartache.py run \   --work-dir ./work/heartache-01 --comfy-root ./h3-local/ComfyUI \   --host http://127.0.0.1:8188 --until-segment 1 --execute

--host http://127.0.0.1:8188 --until-segment 1 --execute

python3 scripts/runtime/heartache.py run \   --work-dir ./work/heartache-01 --comfy-root ./h3-local/ComfyUI \   --host http://127.0.0.1:8188 --until-segment 2 --execute

--host http://127.0.0.1:8188 --until-segment 2 --execute

python3 scripts/runtime/heartache.py run \   --work-dir ./work/heartache-01 --comfy-root ./h3-local/ComfyUI \   --host http://127.0.0.1:8188 --execute

--host http://127.0.0.1:8188 --execute

- Agent 可用 `terminal`、`read_file`、`write_file`，必要时用 `web_search`、`vision_analyze`确认角色输入。所有脚本路径相对于本目录。 - Python、FFmpeg/ffprobe；离线准备可在 macOS/Linux 完成。固定生成环境为 Linux x86_64、Python 3.10、NVIDIA/CUDA。不能把 Mac 上的离线测试当作 H3 原生推理支持。 - 默认**复用用户已有的 Comfy

- Python、FFmpeg/ffprobe；离线准备可在 macOS/Linux 完成。固定生成环境为 Linux x86_64、Python 3.10、NVIDIA/CUDA。不能把 Mac 上的离线测试当作 H3 原生推理支持。 - 默认**复用用户已有的 ComfyUI 与模型**，不要求为了这个模板重装或对齐版本。只有在缺节点、缺模型、版本冲突或推理报错、需要判断该改成什么版本时，才查 [环境参考记录](https://github.com/Shenrui-Ma/

- 默认**复用用户已有的 ComfyUI 与模型**，不要求为了这个模板重装或对齐版本。只有在缺节点、缺模型、版本冲突或推理报错、需要判断该改成什么版本时，才查 [环境参考记录](https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit/tree/667eafddb9f42bb6c72ad27b665def9ada36df45/environments/h3/README.md)，按需取用其中一段。本机 RAM/显存必须足够

[`runtime-assets.json`](runtime-assets.json) 为每项素材记录公开 HTTPS URL、精确字节数、SHA-256、用途及 `download_by_default`。下载源是 [Video ReGen 媒体库](https://huggingface.co/datasets/Shenrui-Ma/video-regen-assets) 的固定 revision，URL 采用 `resolve/<revision>/<path>`，因

`BUNDLE-MANIFEST.json` 记录包含文件及被省略文件的字节数与 SHA-256；省略清单的兼容字段名为 `omitted_downloadable_assets`，**不代表其中每个任意本地二进制都有公开下载源**。素材获取以 `runtime-assets.json` 为准；模型来源、固定版本及位置另见 [环境依赖锁](https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit/tree/667eafddb9

"bytes": 160208343,       "sha256": "e63386ab88bf4a11dc2fd859b0075099229cad44d77366a220c6e0bb0ba0ca0d",       "url": "https://huggingface.co/datasets/Shenrui-Ma/video-regen-assets/resolve/19afd33c5405bcbe1e286cdb35a6c784180e179d/templates/b

"sha256": "e63386ab88bf4a11dc2fd859b0075099229cad44d77366a220c6e0bb0ba0ca0d",       "url": "https://huggingface.co/datasets/Shenrui-Ma/video-regen-assets/resolve/19afd33c5405bcbe1e286cdb35a6c784180e179d/templates/bit-by-bit-heartache/driver

"url": "https://huggingface.co/datasets/Shenrui-Ma/video-regen-assets/resolve/19afd33c5405bcbe1e286cdb35a6c784180e179d/templates/bit-by-bit-heartache/driver/heartache-driver-577f.mp4",

"bytes": 474969,       "sha256": "0da44f53822c7c8ab54fe4b8b089c01c03e89a198e81a2b3cb4d5c8867e4c6c6",       "url": "https://huggingface.co/datasets/Shenrui-Ma/video-regen-assets/resolve/19afd33c5405bcbe1e286cdb35a6c784180e179d/templates/bit-

"sha256": "0da44f53822c7c8ab54fe4b8b089c01c03e89a198e81a2b3cb4d5c8867e4c6c6",       "url": "https://huggingface.co/datasets/Shenrui-Ma/video-regen-assets/resolve/19afd33c5405bcbe1e286cdb35a6c784180e179d/templates/bit-by-bit-heartache/music/

"url": "https://huggingface.co/datasets/Shenrui-Ma/video-regen-assets/resolve/19afd33c5405bcbe1e286cdb35a6c784180e179d/templates/bit-by-bit-heartache/music/music.m4a",

"bytes": 5146149,       "sha256": "97d0d6f5e85b0f0e1b95f5fe8df938402907081e6672c4c0a3f0b4bce87f3bc3",       "url": "https://huggingface.co/datasets/Shenrui-Ma/video-regen-assets/resolve/19afd33c5405bcbe1e286cdb35a6c784180e179d/templates/bit

"sha256": "97d0d6f5e85b0f0e1b95f5fe8df938402907081e6672c4c0a3f0b4bce87f3bc3",       "url": "https://huggingface.co/datasets/Shenrui-Ma/video-regen-assets/resolve/19afd33c5405bcbe1e286cdb35a6c784180e179d/templates/bit-by-bit-heartache/charac

"url": "https://huggingface.co/datasets/Shenrui-Ma/video-regen-assets/resolve/19afd33c5405bcbe1e286cdb35a6c784180e179d/templates/bit-by-bit-heartache/character/yaoguang.png",

"bytes": 59882501,       "sha256": "16806c7e5e3214fd52f65a6d45243ee806805b815e016cb92552c7533f0ebe32",       "url": "https://huggingface.co/datasets/Shenrui-Ma/video-regen-assets/resolve/19afd33c5405bcbe1e286cdb35a6c784180e179d/templates/bi

"sha256": "16806c7e5e3214fd52f65a6d45243ee806805b815e016cb92552c7533f0ebe32",       "url": "https://huggingface.co/datasets/Shenrui-Ma/video-regen-assets/resolve/19afd33c5405bcbe1e286cdb35a6c784180e179d/templates/bit-by-bit-heartache/driver

"url": "https://huggingface.co/datasets/Shenrui-Ma/video-regen-assets/resolve/19afd33c5405bcbe1e286cdb35a6c784180e179d/templates/bit-by-bit-heartache/driver/driver.mp4",

"bytes": 5027044,       "sha256": "5dacb45830e41f852b6ba4751671e8dd64976f0ce570f4cea55ef77430501d09",       "url": "https://huggingface.co/datasets/Shenrui-Ma/video-regen-assets/resolve/19afd33c5405bcbe1e286cdb35a6c784180e179d/templates/bit

"sha256": "5dacb45830e41f852b6ba4751671e8dd64976f0ce570f4cea55ef77430501d09",       "url": "https://huggingface.co/datasets/Shenrui-Ma/video-regen-assets/resolve/19afd33c5405bcbe1e286cdb35a6c784180e179d/templates/bit-by-bit-heartache/clips/

"url": "https://huggingface.co/datasets/Shenrui-Ma/video-regen-assets/resolve/19afd33c5405bcbe1e286cdb35a6c784180e179d/templates/bit-by-bit-heartache/clips/part-01.mp4",

"bytes": 3479889,       "sha256": "e4a8ab9265f617149653979c547b238a4aa8e154e08667c98a86bc4d803afd00",       "url": "https://huggingface.co/datasets/Shenrui-Ma/video-regen-assets/resolve/19afd33c5405bcbe1e286cdb35a6c784180e179d/templates/bit

"sha256": "e4a8ab9265f617149653979c547b238a4aa8e154e08667c98a86bc4d803afd00",       "url": "https://huggingface.co/datasets/Shenrui-Ma/video-regen-assets/resolve/19afd33c5405bcbe1e286cdb35a6c784180e179d/templates/bit-by-bit-heartache/clips/

"url": "https://huggingface.co/datasets/Shenrui-Ma/video-regen-assets/resolve/19afd33c5405bcbe1e286cdb35a6c784180e179d/templates/bit-by-bit-heartache/clips/part-02.mp4",

"bytes": 2696339,       "sha256": "b89312f5954ece2f786b6eccdc934fe5388904dcc3980c7d35a3f0c260161a1c",       "url": "https://huggingface.co/datasets/Shenrui-Ma/video-regen-assets/resolve/19afd33c5405bcbe1e286cdb35a6c784180e179d/templates/bit

"sha256": "b89312f5954ece2f786b6eccdc934fe5388904dcc3980c7d35a3f0c260161a1c",       "url": "https://huggingface.co/datasets/Shenrui-Ma/video-regen-assets/resolve/19afd33c5405bcbe1e286cdb35a6c784180e179d/templates/bit-by-bit-heartache/clips/

"url": "https://huggingface.co/datasets/Shenrui-Ma/video-regen-assets/resolve/19afd33c5405bcbe1e286cdb35a6c784180e179d/templates/bit-by-bit-heartache/clips/part-03.mp4",

"bytes": 4389104,       "sha256": "57ccfef30bd340869cc9f41a797c6c9b6851c1fed270f715750b305cdab04645",       "url": "https://huggingface.co/datasets/Shenrui-Ma/video-regen-assets/resolve/19afd33c5405bcbe1e286cdb35a6c784180e179d/templates/bit

"sha256": "57ccfef30bd340869cc9f41a797c6c9b6851c1fed270f715750b305cdab04645",       "url": "https://huggingface.co/datasets/Shenrui-Ma/video-regen-assets/resolve/19afd33c5405bcbe1e286cdb35a6c784180e179d/templates/bit-by-bit-heartache/clips/

"url": "https://huggingface.co/datasets/Shenrui-Ma/video-regen-assets/resolve/19afd33c5405bcbe1e286cdb35a6c784180e179d/templates/bit-by-bit-heartache/clips/part-04.mp4",

- 节点来源、固定版本、模型哈希和许可说明都在 Toolkit 的[去背景工作流包](https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit/tree/667eafd/workflows/images/rmbg-2-alpha/)（`README.md`、`dependencies.lock.json`）。使用公开`1038lab/ComfyUI-RMBG`仓库的`bd509b4750c81221b938684489c871

- 模型为`1038lab/RMBG-2.0`，revision为`1cd4787601caeb4c8e826dba7ea8e2163b5208df`，放在`ComfyUI/models/RMBG/RMBG-2.0/`。四个文件的URL、SHA256、字节数见该包的 `dependencies.lock.json`。本包不含权重或第三方模型代码。 - 节点GPL-3.0和模型使用权分开确认。先阅读[BRIA官方模型页](https://huggingface.co/briaa

- 节点GPL-3.0和模型使用权分开确认。先阅读[BRIA官方模型页](https://huggingface.co/briaai/RMBG-2.0)及适用许可，确认非商用/商用范围。镜像能下载不表示获得模型授权；官方页面需登录时由用户完成登录，不从镜像绕过授权。

git clone https://github.com/1038lab/ComfyUI-RMBG.git custom_nodes/ComfyUI-RMBG

https://huggingface.co/1038lab/RMBG-2.0/resolve/1cd4787601caeb4c8e826dba7ea8e2163b5208df/config.json

https://huggingface.co/1038lab/RMBG-2.0/resolve/1cd4787601caeb4c8e826dba7ea8e2163b5208df/birefnet.py

https://huggingface.co/1038lab/RMBG-2.0/resolve/1cd4787601caeb4c8e826dba7ea8e2163b5208df/BiRefNet_config.py

https://huggingface.co/1038lab/RMBG-2.0/resolve/1cd4787601caeb4c8e826dba7ea8e2163b5208df/model.safetensors

用 Toolkit 的[去背景包](https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit/tree/667eafd/workflows/images/rmbg-2-alpha/api.template.json)里的参数化API图（`api.template.json`）；本模板不再自带这张图。可用ComfyUI原生节点手动建立等价图：

**RMBG是可选前处理，不是默认生成依赖。** 仅在用户需要去背景时，按[background-removal.md](background-removal.md)使用 Toolkit 的[去背景工作流](https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit/tree/667eafd/workflows/images/rmbg-2-alpha/)，保留原参考、alpha和mask后再运行本helper。本helper不会

python3 scripts/character/prepare_character.py \   --url 'https://OFFICIAL_IMAGE_HOST/character.png' \

--url 'https://OFFICIAL_IMAGE_HOST/character.png' \

当前已提供独立运行入口、完整公开依赖锁与工作流。环境的版本参考见[环境参考记录](https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit/tree/667eafddb9f42bb6c72ad27b665def9ada36df45/environments/h3/README.md)（默认先复用用户已有环境），当前依赖以[环境依赖锁](https://github.com/Shenrui-Ma/shenrui-comfyui-

{   "schema_version": 1,   "repository": "https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit",

"schema_version": 1,   "repository": "https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit",   "environment": "environments/h3",

"repository": "https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit",

"commit": "667eafddb9f42bb6c72ad27b665def9ada36df45",   "guide": "environments/h3/README.md",   "tree_url": "https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit/tree/667eafddb9f42bb6c72ad27b665def9ada36df45/environments/h3",

"guide": "environments/h3/README.md",   "tree_url": "https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit/tree/667eafddb9f42bb6c72ad27b665def9ada36df45/environments/h3",   "dependency_lock": "environments/h3/dependencies.lock.json",

"tree_url": "https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit/tree/667eafddb9f42bb6c72ad27b665def9ada36df45/environments/h3",

"id": "rmbg-2-character-alpha",       "path": "workflows/images/rmbg-2-alpha",       "tree_url": "https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit/tree/667eafddb9f42bb6c72ad27b665def9ada36df45/workflows/images/rmbg-2-alpha",

"path": "workflows/images/rmbg-2-alpha",       "tree_url": "https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit/tree/667eafddb9f42bb6c72ad27b665def9ada36df45/workflows/images/rmbg-2-alpha",       "purpose": "可选的辅助步骤：角色图去背景，输出透明图与蒙版。用户没要求就

"tree_url": "https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit/tree/667eafddb9f42bb6c72ad27b665def9ada36df45/workflows/images/rmbg-2-alpha",

固定基础为1344×768、24fps、20步、res_multistep/simple、BasicGuider、denoise=1。模型与依赖版本参考见[环境参考记录](https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit/tree/667eafddb9f42bb6c72ad27b665def9ada36df45/environments/h3/README.md)（默认复用用户已有环境），角色文字使用[prompt模板

python3 scripts/runtime/heartache.py run \   --work-dir ./work/heartache-01 --comfy-root ./h3-local/ComfyUI \   --host http://127.0.0.1:8188 --until-segment "$N" --execute

--host http://127.0.0.1:8188 --until-segment "$N" --execute

> 历史来源记录：本页描述旧实现/回收阶段。新安装使用 [当前生成路径](generation.md)与[环境说明](https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit/tree/667eafddb9f42bb6c72ad27b665def9ada36df45/environments/h3/README.md)，不需要旧私有Save/Load/Trim节点。

"scope": "historical-recovered-H3-graphs-only; not-current-install-manifest",   "status": "historical-source-record; private Save/Load/Trim are not current dependencies",   "current_dependency_lock": "https://github.com/Shenrui-Ma/shenrui-c

"status": "historical-source-record; private Save/Load/Trim are not current dependencies",   "current_dependency_lock": "https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit/tree/667eafddb9f42bb6c72ad27b665def9ada36df45/environments/h3/dep

"current_dependency_lock": "https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit/tree/667eafddb9f42bb6c72ad27b665def9ada36df45/environments/h3/dependencies.lock.json",

"current_install_guide": "https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit/tree/667eafddb9f42bb6c72ad27b665def9ada36df45/environments/h3/README.md",

"BasicGuider": {       "kind": "core",       "repository": "https://github.com/Comfy-Org/ComfyUI",

"kind": "core",       "repository": "https://github.com/Comfy-Org/ComfyUI",       "python_module": "comfy_extras.nodes_custom_sampler",

"repository": "https://github.com/Comfy-Org/ComfyUI",

"BasicScheduler": {       "kind": "core",       "repository": "https://github.com/Comfy-Org/ComfyUI",

"kind": "core",       "repository": "https://github.com/Comfy-Org/ComfyUI",       "python_module": "comfy_extras.nodes_custom_sampler",

"repository": "https://github.com/Comfy-Org/ComfyUI",

"CLIPLoader": {       "kind": "core",       "repository": "https://github.com/Comfy-Org/ComfyUI",

"kind": "core",       "repository": "https://github.com/Comfy-Org/ComfyUI",       "python_module": "nodes",

"repository": "https://github.com/Comfy-Org/ComfyUI",

"CreateVideo": {       "kind": "core",       "repository": "https://github.com/Comfy-Org/ComfyUI",

"kind": "core",       "repository": "https://github.com/Comfy-Org/ComfyUI",       "python_module": "comfy_extras.nodes_video",

"repository": "https://github.com/Comfy-Org/ComfyUI",

"GetVideoComponents": {       "kind": "core",       "repository": "https://github.com/Comfy-Org/ComfyUI",

"kind": "core",       "repository": "https://github.com/Comfy-Org/ComfyUI",       "python_module": "comfy_extras.nodes_video",

"repository": "https://github.com/Comfy-Org/ComfyUI",

"KSamplerSelect": {       "kind": "core",       "repository": "https://github.com/Comfy-Org/ComfyUI",

"kind": "core",       "repository": "https://github.com/Comfy-Org/ComfyUI",       "python_module": "comfy_extras.nodes_custom_sampler",

"repository": "https://github.com/Comfy-Org/ComfyUI",

"LoadImage": {       "kind": "core",       "repository": "https://github.com/Comfy-Org/ComfyUI",

"kind": "core",       "repository": "https://github.com/Comfy-Org/ComfyUI",       "python_module": "nodes",

"repository": "https://github.com/Comfy-Org/ComfyUI",

"LoadVideo": {       "kind": "core",       "repository": "https://github.com/Comfy-Org/ComfyUI",

"kind": "core",       "repository": "https://github.com/Comfy-Org/ComfyUI",       "python_module": "comfy_extras.nodes_video",

"repository": "https://github.com/Comfy-Org/ComfyUI",

"kind": "custom",       "package": "H3MotionContextOfficial",       "repository": "https://github.com/NikoDemon80/ComfyUI-H3-Motion-Context",

"package": "H3MotionContextOfficial",       "repository": "https://github.com/NikoDemon80/ComfyUI-H3-Motion-Context",       "commit": "f80e36bc1d7887a143b12e6645313fd6b9cd2aee",

"repository": "https://github.com/NikoDemon80/ComfyUI-H3-Motion-Context",

"MiniMaxH3ReferenceToVideo": {       "kind": "core",       "repository": "https://github.com/Comfy-Org/ComfyUI",

"kind": "core",       "repository": "https://github.com/Comfy-Org/ComfyUI",       "python_module": "comfy_extras.nodes_minimax_h3",

"repository": "https://github.com/Comfy-Org/ComfyUI",

"RandomNoise": {       "kind": "core",       "repository": "https://github.com/Comfy-Org/ComfyUI",

"kind": "core",       "repository": "https://github.com/Comfy-Org/ComfyUI",       "python_module": "comfy_extras.nodes_custom_sampler",

"repository": "https://github.com/Comfy-Org/ComfyUI",

"SamplerCustomAdvanced": {       "kind": "core",       "repository": "https://github.com/Comfy-Org/ComfyUI",

"kind": "core",       "repository": "https://github.com/Comfy-Org/ComfyUI",       "python_module": "comfy_extras.nodes_custom_sampler",

"repository": "https://github.com/Comfy-Org/ComfyUI",

"SaveVideo": {       "kind": "core",       "repository": "https://github.com/Comfy-Org/ComfyUI",

"kind": "core",       "repository": "https://github.com/Comfy-Org/ComfyUI",       "python_module": "comfy_extras.nodes_video",

"repository": "https://github.com/Comfy-Org/ComfyUI",

"UNETLoader": {       "kind": "core",       "repository": "https://github.com/Comfy-Org/ComfyUI",

"kind": "core",       "repository": "https://github.com/Comfy-Org/ComfyUI",       "python_module": "nodes",

"repository": "https://github.com/Comfy-Org/ComfyUI",

"VAEDecode": {       "kind": "core",       "repository": "https://github.com/Comfy-Org/ComfyUI",

"kind": "core",       "repository": "https://github.com/Comfy-Org/ComfyUI",       "python_module": "nodes",

"repository": "https://github.com/Comfy-Org/ComfyUI",

"VAEDecodeAudio": {       "kind": "core",       "repository": "https://github.com/Comfy-Org/ComfyUI",

"kind": "core",       "repository": "https://github.com/Comfy-Org/ComfyUI",       "python_module": "comfy_extras.nodes_audio",

"repository": "https://github.com/Comfy-Org/ComfyUI",

"VAELoader": {       "kind": "core",       "repository": "https://github.com/Comfy-Org/ComfyUI",

"kind": "core",       "repository": "https://github.com/Comfy-Org/ComfyUI",       "python_module": "nodes",

"repository": "https://github.com/Comfy-Org/ComfyUI",

**本页及 [node-sources.json](node-sources.json) 的 `node_sources` 是历史恢复图的来源记录，不是当前安装清单。当前运行图由 [workflow.lock.json](workflow.lock.json) 锁定到 toolkit 的 Core 续接包，其依赖以 [环境依赖锁](https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit/tree/667eafddb9f42b

- 续接：两路 Core `LoadLatent` 读回，由 `LTXVConcatAVLatent` 重组后送入 `MiniMaxH3MotionContext.context_latent`。不能只保存视频 latent 或用末帧代替联合 AV 续接。 - 裁切：当前图及运行层负责发布帧/音频裁切，不调用历史 `TrimMiniMaxH3MotionContext`。这是当前公开替代实现，非历史私有缓存格式的字节等价承诺。 - 唯一自定义包为 [NikoDemon80/C

- 裁切：当前图及运行层负责发布帧/音频裁切，不调用历史 `TrimMiniMaxH3MotionContext`。这是当前公开替代实现，非历史私有缓存格式的字节等价承诺。 - 唯一自定义包为 [NikoDemon80/ComfyUI-H3-Motion-Context](https://github.com/NikoDemon80/ComfyUI-H3-Motion-Context)，固定 commit `f80e36bc1d7887a143b12e6645313fd6b9

- 唯一自定义包为 [NikoDemon80/ComfyUI-H3-Motion-Context](https://github.com/NikoDemon80/ComfyUI-H3-Motion-Context)，固定 commit `f80e36bc1d7887a143b12e6645313fd6b9cd2aee`，GPL-3.0。`H3MotionContextOfficial` 只是目录名，不表示官方 Core。

对应 [Issue #1](https://github.com/Shenrui-Ma/video-regen-recipes/issues/1)。本页补充原版运行证据，供已经用其他节点跑通生成的复现者对照。节点同名、参数接近、能够生成，都不证明上下文行为或输入相同。

- [实际提示词原文](../prompts/yaoguang.historical.txt)：四段逐字相同，不是占位模板展开后的四套新词。 - [原生续接语义](native-continuation.md)：AV latent、音频换算、窗口、Trim及两种MotionContext的区别。 - [历史577帧驱动母片](https://huggingface.co/datasets/Shenrui-Ma/video-regen-assets/resolve/19afd3

- [原生续接语义](native-continuation.md)：AV latent、音频换算、窗口、Trim及两种MotionContext的区别。 - [历史577帧驱动母片](https://huggingface.co/datasets/Shenrui-Ma/video-regen-assets/resolve/19afd33c5405bcbe1e286cdb35a6c784180e179d/templates/bit-by-bit-heartache/drive

- [历史577帧驱动母片](https://huggingface.co/datasets/Shenrui-Ma/video-regen-assets/resolve/19afd33c5405bcbe1e286cdb35a6c784180e179d/templates/bit-by-bit-heartache/driver/heartache-driver-577f.mp4)：160,208,343字节，SHA-256 `e63386ab88bf4a11dc2fd859b0

> 历史来源记录：本页描述旧实现/回收阶段。新安装使用 [当前生成路径](generation.md)与[环境说明](https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit/tree/667eafddb9f42bb6c72ad27b665def9ada36df45/environments/h3/README.md)，不需要旧私有Save/Load/Trim节点。

当时的串行控制器、四份重建 API 图及部分 History 回读仍有保存。当前运行图不再放在本仓库：Toolkit 的 [H3 Core AV latent 续接包](https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit/tree/af3dea3d05450eb2910f43f795a2eb6f346d06dc/workflows/h3/ref2va-core-continuation) 提供首段／续段两份参数化图，具体

{   "schema_version": 2,   "repository": "https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit",

"schema_version": 2,   "repository": "https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit",   "workflow_id": "h3-ref2va-core-continuation",

"repository": "https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit",

"path": "workflows/h3/ref2va-core-continuation/first.api.template.json",       "sha256": "e15319f545f1f3a363e9f8a2de82ecdbdfeb349d8c0d8522770b1060796ab235",       "url": "https://raw.githubusercontent.com/Shenrui-Ma/shenrui-comfyui-toolkit/

"sha256": "e15319f545f1f3a363e9f8a2de82ecdbdfeb349d8c0d8522770b1060796ab235",       "url": "https://raw.githubusercontent.com/Shenrui-Ma/shenrui-comfyui-toolkit/af3dea3d05450eb2910f43f795a2eb6f346d06dc/workflows/h3/ref2va-core-continuation/

"url": "https://raw.githubusercontent.com/Shenrui-Ma/shenrui-comfyui-toolkit/af3dea3d05450eb2910f43f795a2eb6f346d06dc/workflows/h3/ref2va-core-continuation/first.api.template.json"

"path": "workflows/h3/ref2va-core-continuation/continue.api.template.json",       "sha256": "41bcbb34f84bd1a8bdd489cf365c4ad18b74dcc0a459e0c3337bcacbb22dcd24",       "url": "https://raw.githubusercontent.com/Shenrui-Ma/shenrui-comfyui-toolk

"sha256": "41bcbb34f84bd1a8bdd489cf365c4ad18b74dcc0a459e0c3337bcacbb22dcd24",       "url": "https://raw.githubusercontent.com/Shenrui-Ma/shenrui-comfyui-toolkit/af3dea3d05450eb2910f43f795a2eb6f346d06dc/workflows/h3/ref2va-core-continuation/

"url": "https://raw.githubusercontent.com/Shenrui-Ma/shenrui-comfyui-toolkit/af3dea3d05450eb2910f43f795a2eb6f346d06dc/workflows/h3/ref2va-core-continuation/continue.api.template.json"

"asset_id": "heartache.driver.mother-577f",     "library_revision": "19afd33c5405bcbe1e286cdb35a6c784180e179d",     "download_url": "https://huggingface.co/datasets/Shenrui-Ma/video-regen-assets/resolve/19afd33c5405bcbe1e286cdb35a6c784180e1

"library_revision": "19afd33c5405bcbe1e286cdb35a6c784180e179d",     "download_url": "https://huggingface.co/datasets/Shenrui-Ma/video-regen-assets/resolve/19afd33c5405bcbe1e286cdb35a6c784180e179d/templates/bit-by-bit-heartache/driver/hearta

"download_url": "https://huggingface.co/datasets/Shenrui-Ma/video-regen-assets/resolve/19afd33c5405bcbe1e286cdb35a6c784180e179d/templates/bit-by-bit-heartache/driver/heartache-driver-577f.mp4"

#!/usr/bin/env python3

opener = urllib.request.build_opener(urllib.request.ProxyHandler({}), PublicRedirect())     return opener.open(urllib.request.Request(url, headers={'User-Agent': 'HeartacheCharacter/1'}), timeout=30)

return opener.open(urllib.request.Request(url, headers={'User-Agent': 'HeartacheCharacter/1'}), timeout=30)

#!/usr/bin/env python3

with source.open('rb') as f: shutil.copyfileobj(f,out,1024*1024)             else:                 with urllib.request.urlopen(row['url'],timeout=60) as f:

else:                 with urllib.request.urlopen(row['url'],timeout=60) as f:                     size=0

with urllib.request.urlopen(row['url'],timeout=60) as f:

#!/usr/bin/env python3

[Credential-related source omitted]

#!/usr/bin/env python3

#!/usr/bin/env python3

#!/usr/bin/env python3

Executable file

if target.is_file() and hashlib.sha256(target.read_bytes()).hexdigest() == entry['sha256']:         return json.loads(target.read_text(encoding='utf-8'))     with urllib.request.urlopen(entry['url'], timeout=60) as response:

return json.loads(target.read_text(encoding='utf-8'))     with urllib.request.urlopen(entry['url'], timeout=60) as response:         data = response.read()

with urllib.request.urlopen(entry['url'], timeout=60) as response:

#!/usr/bin/env python3

r.add_argument('--work-dir', required=True)     r.add_argument('--comfy-root')     r.add_argument('--host', default='http://127.0.0.1:8188')

r.add_argument('--comfy-root')     r.add_argument('--host', default='http://127.0.0.1:8188')     r.add_argument('--execute', action='store_true')

r.add_argument('--host', default='http://127.0.0.1:8188')

Executable file

Executable file

#!/usr/bin/env python3

import errno import hashlib import http.client

import hashlib import http.client import json

import http.client

try:                 data = None if body is None else json.dumps(body, allow_nan=False).encode()                 req = urllib.request.Request(self.host + path, data=data,

data = None if body is None else json.dumps(body, allow_nan=False).encode()                 req = urllib.request.Request(self.host + path, data=data,                                              headers={"Content-Type": "application/json"},

req = urllib.request.Request(self.host + path, data=data,

try:                             result = json.load(error)                         except (OSError, ValueError, http.client.HTTPException):

result = json.load(error)                         except (OSError, ValueError, http.client.HTTPException):                             raise RunnerError("HTTP_IO_FAILED") from None

except (OSError, ValueError, http.client.HTTPException):

if attempt == 2:                     raise RunnerError("HTTP_IO_FAILED") from None             except (OSError, ValueError, urllib.error.URLError, http.client.HTTPException):

raise RunnerError("HTTP_IO_FAILED") from None             except (OSError, ValueError, urllib.error.URLError, http.client.HTTPException):                 if method == "POST" or attempt == 2:

except (OSError, ValueError, urllib.error.URLError, http.client.HTTPException):

当前[环境依赖锁](https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit/tree/667eafddb9f42bb6c72ad27b665def9ada36df45/environments/h3/dependencies.lock.json)提供H3模型权重的固定来源、大小和SHA，安装脚本按固定版本获取公开节点。权重、私人latent和成片不内嵌轻量包。历史剪辑素材仅用于用户明确授权的重剪；新角色仍需实际生成，不能仅凭

技术依赖：[MiniMax H3](https://github.com/MiniMax-AI/MiniMax-H3)、[模型卡](https://huggingface.co/MiniMaxAI/MiniMax-H3)、[ComfyUI](https://github.com/Comfy-Org/ComfyUI)、[FFmpeg](https://ffmpeg.org/)。可选RTX VSR和RIFE依其实际实现与许可使用。

Executable file

Executable file

Image.new('RGB', (16, 24), 'red').save(source)             out = Path(tmp) / 'prepared'             with patch('urllib.request.urlopen', side_effect=AssertionError('offline')):

out = Path(tmp) / 'prepared'             with patch('urllib.request.urlopen', side_effect=AssertionError('offline')):                 result = m.main(['--image', str(source), '--out-dir', str(out)])

with patch('urllib.request.urlopen', side_effect=AssertionError('offline')):

original = source.read_bytes()             out = Path(tmp) / 'white'             with patch('urllib.request.urlopen', side_effect=AssertionError('offline')):

out = Path(tmp) / 'white'             with patch('urllib.request.urlopen', side_effect=AssertionError('offline')):                 try:

with patch('urllib.request.urlopen', side_effect=AssertionError('offline')):

with self.subTest(color=color), patch.object(m, 'open_public', side_effect=AssertionError('must not download')):                     with self.assertRaisesRegex(ValueError, '#RRGGBB'):                         m.main(['--url', 'https://examp

with self.assertRaisesRegex(ValueError, '#RRGGBB'):                         m.main(['--url', 'https://example.org/image.png', '--background', color, '--out-dir', str(out)])                     self.assertFalse(out.exists())

m.main(['--url', 'https://example.org/image.png', '--background', color, '--out-dir', str(out)])

out = Path(tmp) / 'url'             with patch.object(m, 'open_public', return_value=io.BytesIO(data)):                 m.main(['--url', 'https://example.org/character.png', '--sha256', hashlib.sha256(data).hexdigest(), '--out-dir', str(out

with patch.object(m, 'open_public', return_value=io.BytesIO(data)):                 m.main(['--url', 'https://example.org/character.png', '--sha256', hashlib.sha256(data).hexdigest(), '--out-dir', str(out)])             p = json.loads((out 

m.main(['--url', 'https://example.org/character.png', '--sha256', hashlib.sha256(data).hexdigest(), '--out-dir', str(out)])

with patch.object(m, 'open_public', return_value=io.BytesIO(data)):                 with self.assertRaises(ValueError):                     m.main(['--url', 'https://example.org/character.png', '--sha256', '0' * 64, '--out-dir', str(bad)])

with self.assertRaises(ValueError):                     m.main(['--url', 'https://example.org/character.png', '--sha256', '0' * 64, '--out-dir', str(bad)])             self.assertFalse((bad / 'character.png').exists())

m.main(['--url', 'https://example.org/character.png', '--sha256', '0' * 64, '--out-dir', str(bad)])

m = load_module()         self.assertTrue(hasattr(m, 'validate_public_url'), 'public URL validation missing')         for url in ('file:///etc/passwd', 'http://127.0.0.1/x', 'https://user:pass@example.org/x', 'http://[::1]/x'):

self.assertTrue(hasattr(m, 'validate_public_url'), 'public URL validation missing')         for url in ('file:///etc/passwd', 'http://127.0.0.1/x', 'https://user:pass@example.org/x', 'http://[::1]/x'):             with self.subTest(url=url)

for url in ('file:///etc/passwd', 'http://127.0.0.1/x', 'https://user:pass@example.org/x', 'http://[::1]/x'):

with patch.object(m, 'open_public', return_value=io.BytesIO(b'not an image')):                 with self.assertRaises(OSError):                     m.main(['--url', 'https://example.org/image.png', '--out-dir', str(out)])

with self.assertRaises(OSError):                     m.main(['--url', 'https://example.org/image.png', '--out-dir', str(out)])             self.assertFalse((out / 'character.png').exists())

m.main(['--url', 'https://example.org/image.png', '--out-dir', str(out)])             self.assertFalse((out / 'character.png').exists())             self.assertFalse((out / 'download.part').exists(), 'failed decode leaves .part')

Executable file

Executable file

m=load('fetch_assets'); self.assertIsNotNone(m,'asset acquisition helper missing')         data=b'offline-test-only'         row={'path':'reference/driver.mp4','bytes':len(data),'sha256':hashlib.sha256(data).hexdigest(),'url':'https://examp

data=b'offline-test-only'         row={'path':'reference/driver.mp4','bytes':len(data),'sha256':hashlib.sha256(data).hexdigest(),'url':'https://example.invalid/driver.mp4','role':'motion_reference'}         with tempfile.TemporaryDirectory(

row={'path':'reference/driver.mp4','bytes':len(data),'sha256':hashlib.sha256(data).hexdigest(),'url':'https://example.invalid/driver.mp4','role':'motion_reference'}

p=Path(td); cache=p/'cache'; cache.mkdir(); (cache/'driver.mp4').write_bytes(data)             out=p/'assets'             with patch('urllib.request.urlopen',side_effect=AssertionError('network forbidden')):

out=p/'assets'             with patch('urllib.request.urlopen',side_effect=AssertionError('network forbidden')):                 result=m.acquire(out,row,cache)

with patch('urllib.request.urlopen',side_effect=AssertionError('network forbidden')):

with tempfile.TemporaryDirectory() as td:             p=Path(td)/'new'             for path,url in [('../escape','https://example.invalid/a'),('/absolute','https://example.invalid/a'),('x','http://example.invalid/a')]:

p=Path(td)/'new'             for path,url in [('../escape','https://example.invalid/a'),('/absolute','https://example.invalid/a'),('x','http://example.invalid/a')]:                 with self.assertRaises(ValueError): m.acquire(p,{'path':pat

for path,url in [('../escape','https://example.invalid/a'),('/absolute','https://example.invalid/a'),('x','http://example.invalid/a')]:

body=b'offline manual fallback fixture'         return {'path':'default/music.m4a','bytes':len(body),'sha256':hashlib.sha256(body).hexdigest(),                 'url':'https://example.invalid/music.m4a','role':'soundtrack','download_by_defau

'url':'https://example.invalid/music.m4a','role':'soundtrack','download_by_default':True}

with tempfile.TemporaryDirectory() as td:             p=Path(td); manifest=p/'manifest.json'; manifest.write_text(json.dumps({'assets':[row]}))             with patch('urllib.request.urlopen',side_effect=AssertionError('network forbidden'))

p=Path(td); manifest=p/'manifest.json'; manifest.write_text(json.dumps({'assets':[row]}))             with patch('urllib.request.urlopen',side_effect=AssertionError('network forbidden')):                 result=subprocess.run([sys.executabl

with patch('urllib.request.urlopen',side_effect=AssertionError('network forbidden')):

body=b'offline manual fallback fixture'         row={'path':'default/music.m4a','bytes':len(body),'sha256':hashlib.sha256(body).hexdigest(),              'url':'https://127.0.0.1:9/music.m4a','role':'soundtrack','download_by_default':True}

row={'path':'default/music.m4a','bytes':len(body),'sha256':hashlib.sha256(body).hexdigest(),              'url':'https://127.0.0.1:9/music.m4a','role':'soundtrack','download_by_default':True}         with tempfile.TemporaryDirectory() as td

'url':'https://127.0.0.1:9/music.m4a','role':'soundtrack','download_by_default':True}

Executable file

[Credential-related source omitted]

Executable file

Executable file

Executable file

Executable file

host = 'http://127.0.0.1:8188'; interval = 0

Executable file

for row in lock['graphs']:             self.assertEqual(len(row['sha256']), 64)             self.assertTrue(row['url'].startswith('https://raw.githubusercontent.com/Shenrui-Ma/shenrui-comfyui-toolkit/'))

self.assertEqual(len(row['sha256']), 64)             self.assertTrue(row['url'].startswith('https://raw.githubusercontent.com/Shenrui-Ma/shenrui-comfyui-toolkit/'))             self.assertIn(lock['commit'], row['url'])

self.assertTrue(row['url'].startswith('https://raw.githubusercontent.com/Shenrui-Ma/shenrui-comfyui-toolkit/'))

Executable file

Executable file

Executable file

r = importlib.import_module('run_segment')         class API:             host = 'http://localhost:8188'

class API:             host = 'http://localhost:8188'             interval = 0

host = 'http://localhost:8188'

Executable file

Executable file

- [shenrui-comfyui-toolkit](https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit)：H3 与图像工作流的维护位置 - [Shenrui-Ma/video-regen-assets](https://huggingface.co/datasets/Shenrui-Ma/video-regen-assets)：公开素材库；参考视频用[参考视频获取 Skill](../../../skills/ref

- [Shenrui-Ma/video-regen-assets](https://huggingface.co/datasets/Shenrui-Ma/video-regen-assets)：公开素材库；参考视频用[参考视频获取 Skill](../../../skills/reference-video-fetch/SKILL.md)按需取到本机

"page_url": "https://www.bilibili.com/video/BV1LQ7j6WEyD/",

具体两次采样图可从 [Toolkit](https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit/tree/main/workflows/images/sdxl-two-pass-cowboy-shot)获取，按[固定版本解析](../../../../skills/comfyui-reference-images/references/toolkit.md)填写模型与提示词。

- [MiniMax H3](https://github.com/MiniMax-AI/MiniMax-H3) · [模型卡与许可](https://huggingface.co/MiniMaxAI/MiniMax-H3) · [提示词规范](https://github.com/MiniMax-AI/MiniMax-H3/tree/main/skills/h3-prompt-writing)

- [ComfyUI](https://github.com/Comfy-Org/ComfyUI)、[RIFE](https://github.com/hzwer/ECCV2022-RIFE)、[FFmpeg](https://ffmpeg.org/)

- 《🐋DeepSeek不是罪过~🐋》 · 作者：**uulumine゛** · [B 站 BV1LQ7j6WEyD](https://www.bilibili.com/video/BV1LQ7j6WEyD/) · 56.0 秒 · 2026-06-26 投稿。 - 这是第三方作品，仅作为动作与节奏参考：用户应当换成自己的参考视频或自备文件。本仓库不分发它，也不因为引用而获得任何授权。 - 获取方式见[参考视频获取 Skill](../../../skills/referen

- **ComfyUI 与所选生图模型**（SDXL、Anima 等）：依各自许可；模型本机已有优先，缺模型时由[参考图 Skill](../../../skills/comfyui-reference-images/SKILL.md)按发布页固定版本查找与核验。本模板不附带权重。 - **共用能力**：[ComfyUI 参考图 Skill](../../../skills/comfyui-reference-images/SKILL.md)、[角色合集剪辑 Skill](.

- **共用能力**：[ComfyUI 参考图 Skill](../../../skills/comfyui-reference-images/SKILL.md)、[角色合集剪辑 Skill](../../../skills/character-showcase-editing/SKILL.md)，以及 [ComfyUI Toolkit](https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit) 中的工作流。仓库 MIT 覆

- [shenrui-comfyui-toolkit](https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit)：固定版本的图片与视频工作流

- [Shenrui-Ma/video-regen-assets](https://huggingface.co/datasets/Shenrui-Ma/video-regen-assets)：公开素材库；本模板不附带媒体

生图依赖采用 [ComfyUI](https://github.com/Comfy-Org/ComfyUI) 与所选模型各自的许可；工作流从 [配套库](https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit) 固定版本获取。新用户自己的角色图、模型与配乐不随模板提供。

需要：可调用 Skill 的 Agent、本地 ComfyUI 和 MiniMax H3、Python + Pillow、FFmpeg。参考图模型会按角色需求匹配；远程 ComfyUI 走独立连接配置。具体工作流由 [ComfyUI Toolkit](https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit) 提供。

- [shenrui-comfyui-toolkit](https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit)：`h3-i2v-live-portrait` 等固定版本工作流，固定 revision 与哈希见 [workflows/toolkit-source.json](workflows/toolkit-source.json)

- [Shenrui-Ma/video-regen-assets](https://huggingface.co/datasets/Shenrui-Ma/video-regen-assets)：公开素材库；本模板不附带媒体

"image_backend": "comfyui",   "video_workflow_id": "h3-i2v-live-portrait",   "comfyui_base_url": "http://127.0.0.1:8188",

"video_workflow_id": "h3-i2v-live-portrait",   "comfyui_base_url": "http://127.0.0.1:8188",   "shot_plan": "eight-shot-plan.json",

"comfyui_base_url": "http://127.0.0.1:8188",

工作流 ID：`h3-i2v-live-portrait`。具体图在 [Toolkit](https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit) 的 `workflows/h3/i2v-live-portrait/`；通过 Recipes 的[固定版本解析器](../../../../skills/comfyui-reference-images/references/toolkit.md)获得图和 SHA-256，按图随

{   "repository": "https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit",

"repository": "https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit",

- [shenrui-comfyui-toolkit](https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit)：`h3-i2v-live-portrait` 等固定版本工作流，固定 revision 与哈希见 [workflows/toolkit-source.json](workflows/toolkit-source.json)

- [Shenrui-Ma/video-regen-assets](https://huggingface.co/datasets/Shenrui-Ma/video-regen-assets)：公开素材库；本模板的示例配置不附带媒体

{   "repository": "https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit",

"repository": "https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit",

- **FFmpeg、Python、Pillow 等依赖**：依各自许可。 - **示例素材**：`assets/reference.png` 与 `examples/preview.mp4` 是作者自有的示例输入与渲染结果，随模板分发，哈希见 [assets/manifest.json](assets/manifest.json)。用他人作品作参考图时请自行确认授权，不要把第三方插画当作可分发素材带进来。 - **灵感来源**：[Hope麻匪《再给 GPT6 Astra 一

- **示例素材**：`assets/reference.png` 与 `examples/preview.mp4` 是作者自有的示例输入与渲染结果，随模板分发，哈希见 [assets/manifest.json](assets/manifest.json)。用他人作品作参考图时请自行确认授权，不要把第三方插画当作可分发素材带进来。 - **灵感来源**：[Hope麻匪《再给 GPT6 Astra 一点时间》](https://www.bilibili.com/video/B

- **灵感来源**：[Hope麻匪《再给 GPT6 Astra 一点时间》](https://www.bilibili.com/video/BV1azYW6zEXy) 仅作署名引用，不代表获得该作品的任何授权。

模板实现：**Shenrui Ma（四倍体果蝇）**。灵感来源：[Hope麻匪《再给 GPT6 Astra 一点时间》](https://www.bilibili.com/video/BV1azYW6zEXy)。[来源与验证](sources.md)

- [shenrui-comfyui-toolkit](https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit)：本仓库其它模板使用的固定版本图像与视频工作流；本模板不使用模型推理

- [Shenrui-Ma/video-regen-assets](https://huggingface.co/datasets/Shenrui-Ma/video-regen-assets)：公开素材库；本模板的示例素材体积与哈希见 [assets/manifest.json](assets/manifest.json)

const status=document.getElementById('load-status'),picker=document.getElementById('svg-files');  const expected=JSON.parse(document.getElementById('art-manifest').textContent);  const ns='http://www.w3.org/2000/svg';

const expected=JSON.parse(document.getElementById('art-manifest').textContent);  const ns='http://www.w3.org/2000/svg';  async function parse(bytes,name){

const ns='http://www.w3.org/2000/svg';

let bytes;   if(files){if(!files.has(name))throw Error('请同时选择两个 SVG 文件，刷新页面后重试。');bytes=await files.get(name).arrayBuffer();}   else {const response=await fetch(name);if(!response.ok)throw Error('无法读取 '+name);bytes=await response.arrayBuffe

if(files){if(!files.has(name))throw Error('请同时选择两个 SVG 文件，刷新页面后重试。');bytes=await files.get(name).arrayBuffer();}   else {const response=await fetch(name);if(!response.ok)throw Error('无法读取 '+name);bytes=await response.arrayBuffer();}   roots

else {const response=await fetch(name);if(!response.ok)throw Error('无法读取 '+name);bytes=await response.arrayBuffer();}

<svg xmlns="http://www.w3.org/2000/svg" id="art" viewBox="0 0 1536 1024" aria-label="SVG临摹重绘"><g id="construction" opacity="0"><polygon id="construction-seated-coat" points="1080,350 1230,335 1290,490 1280,830 1410,1024 1070,1024 951,784 95

Executable file

Executable file

Executable file

import math,re,xml.etree.ElementTree as ET NS='{http://www.w3.org/2000/svg}'

NS='{http://www.w3.org/2000/svg}'

Executable file

svg=read_contours(args.svg.read_text(),W,H,[name for name,_ in regions]);ns='{http://www.w3.org/2000/svg}'

points=' '.join(f'{x},{y}' for x,y in poly)   sketch.append(f'<polygon id="construction-{name}" points="{points}" fill="none" stroke="#8d9bb5" stroke-width="2"/>') art=f'''<svg id="art" xmlns="http://www.w3.org/2000/svg" viewBox="0 0 {W} {H

sketch.append(f'<polygon id="construction-{name}" points="{points}" fill="none" stroke="#8d9bb5" stroke-width="2"/>') art=f'''<svg id="art" xmlns="http://www.w3.org/2000/svg" viewBox="0 0 {W} {H}" aria-label="SVG临摹重绘"><g id="construction" o

art=f'''<svg id="art" xmlns="http://www.w3.org/2000/svg" viewBox="0 0 {W} {H}" aria-label="SVG临摹重绘"><g id="construction" opacity="0">{''.join(sketch)}</g><g id="flat-colors" shape-rendering="crispEdges" fill-rule="evenodd">{''.join(flats)}<

#!/usr/bin/env python3

#!/usr/bin/env node

Executable file

NS='http://www.w3.org/2000/svg'

Executable file

header=f'<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 {W} {H}" width="{W}" height="{H}" role="img" aria-label="Reference illustration redrawn as native color-region contours">'

灵感来源：Bilibili **Hope麻匪**，[《再给 GPT6 Astra 一点时间》](https://www.bilibili.com/video/BV1azYW6zEXy)。已核对作者和视频标题；没有转载其视频或声称拿到了其SVG源码。

Executable file

});   await new Promise(resolve=>server.listen(0,'127.0.0.1',resolve));   try{await page.goto(`http://127.0.0.1:${server.address().port}/demo.html`);await page.waitForFunction(()=>window.replay,{},{timeout:120000});assert.equal(await page.l

await new Promise(resolve=>server.listen(0,'127.0.0.1',resolve));   try{await page.goto(`http://127.0.0.1:${server.address().port}/demo.html`);await page.waitForFunction(()=>window.replay,{},{timeout:120000});assert.equal(await page.locator

try{await page.goto(`http://127.0.0.1:${server.address().port}/demo.html`);await page.waitForFunction(()=>window.replay,{},{timeout:120000});assert.equal(await page.locator('#svg-files').isHidden(),true);}

Executable file

with self.assertRaises(ValueError):read_regions(p,2,1,b'f')  def valid(self):   return '<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 2 1"><g id="complete-vector-art"><g data-region="background"><path fill="#aabbcc" d="M0 0h2v1h-2z"/

def valid(self):   return '<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 2 1"><g id="complete-vector-art"><g data-region="background"><path fill="#aabbcc" d="M0 0h2v1h-2z"/></g></g></svg>'  def test_vector_validation(self):

return '<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 2 1"><g id="complete-vector-art"><g data-region="background"><path fill="#aabbcc" d="M0 0h2v1h-2z"/></g></g></svg>'

for marker in ['<canvas','<img','<image','<foreignObject','data:image','<script src=','http://','https://']:    if marker=='http://':     self.assertNotIn(marker,text.replace('http://www.w3.org/2000/svg',''))

if marker=='http://':     self.assertNotIn(marker,text.replace('http://www.w3.org/2000/svg',''))    else:self.assertNotIn(marker,text)

self.assertNotIn(marker,text.replace('http://www.w3.org/2000/svg',''))

root=Path(tmp);source=root/'source.svg';source.write_text(self.valid().replace('<path fill=','<path id="tone-1" fill='))    path='<path id="tone-1" fill="#aabbcc" d="M0 0h2v1h-2z"/>'    art='<svg id="art" xmlns="http://www.w3.org/2000/svg" 

path='<path id="tone-1" fill="#aabbcc" d="M0 0h2v1h-2z"/>'    art='<svg id="art" xmlns="http://www.w3.org/2000/svg" viewBox="0 0 2 1"><g id="construction"/><g id="flat-colors"/><g id="ink"/><g id="final-contours"><g data-stage="shadow">'+pa

art='<svg id="art" xmlns="http://www.w3.org/2000/svg" viewBox="0 0 2 1"><g id="construction"/><g id="flat-colors"/><g id="ink"/><g id="final-contours"><g data-stage="shadow">'+path+'</g></g></svg>'

Executable file

const status=document.getElementById('load-status'),picker=document.getElementById('svg-files');  const expected=JSON.parse(document.getElementById('art-manifest').textContent);  const ns='http://www.w3.org/2000/svg';

const expected=JSON.parse(document.getElementById('art-manifest').textContent);  const ns='http://www.w3.org/2000/svg';  async function parse(bytes,name){

const ns='http://www.w3.org/2000/svg';

let bytes;   if(files){if(!files.has(name))throw Error('请同时选择两个 SVG 文件，刷新页面后重试。');bytes=await files.get(name).arrayBuffer();}   else {const response=await fetch(name);if(!response.ok)throw Error('无法读取 '+name);bytes=await response.arrayBuffe

if(files){if(!files.has(name))throw Error('请同时选择两个 SVG 文件，刷新页面后重试。');bytes=await files.get(name).arrayBuffer();}   else {const response=await fetch(name);if(!response.ok)throw Error('无法读取 '+name);bytes=await response.arrayBuffer();}   roots

else {const response=await fetch(name);if(!response.ok)throw Error('无法读取 '+name);bytes=await response.arrayBuffer();}

Executable file

- [𝑺𝑻𝑨𝑹𝑩𝑶𝒀 高松灯](https://www.bilibili.com/video/BV1BPgq6hEey/)

- [𝑺𝑻𝑨𝑹𝑩𝑶𝒀 高松灯](https://www.bilibili.com/video/BV1BPgq6hEey/)

- [𝑺𝑻𝑨𝑹𝑩𝑶𝒀 高松灯](https://www.bilibili.com/video/BV1BPgq6hEey/)

- [𝑩𝒊𝒍𝒍𝒊𝒆 𝑱𝒆𝒂𝒏 —— 千早爱音](https://www.bilibili.com/video/BV1dQKY6ME3t/)

- [𝑩𝒊𝒍𝒍𝒊𝒆 𝑱𝒆𝒂𝒏 —— 千早爱音](https://www.bilibili.com/video/BV1dQKY6ME3t/)

- [𝑩𝒊𝒍𝒍𝒊𝒆 𝑱𝒆𝒂𝒏 —— 千早爱音](https://www.bilibili.com/video/BV1dQKY6ME3t/)

- [shenrui-comfyui-toolkit](https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit)：本仓库其它模板使用的固定版本工作流；本模板是独立渲染器，不调用模型

- [shenrui-comfyui-toolkit](https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit)：本仓库其它模板使用的固定版本工作流；本模板是独立渲染器，不调用模型

- [shenrui-comfyui-toolkit](https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit)：本仓库其它模板使用的固定版本工作流；本模板是独立渲染器，不调用模型

- [Shenrui-Ma/video-regen-assets](https://huggingface.co/datasets/Shenrui-Ma/video-regen-assets)：公开素材库；母带、图片与字体由用户自备

- [Shenrui-Ma/video-regen-assets](https://huggingface.co/datasets/Shenrui-Ma/video-regen-assets)：公开素材库；母带、图片与字体由用户自备

- [Shenrui-Ma/video-regen-assets](https://huggingface.co/datasets/Shenrui-Ma/video-regen-assets)：公开素材库；母带、图片与字体由用户自备

"related_templates": [{"id": "girls-band-ddsp-cover", "guide": "../../singing-cover/girls-band-ddsp/README.md", "optional": true}],   "examples": [     {"title": "𝑺𝑻𝑨𝑹𝑩𝑶𝒀 高松灯", "url": "https://www.bilibili.com/video/BV1BPgq6hEey/"},

"examples": [     {"title": "𝑺𝑻𝑨𝑹𝑩𝑶𝒀 高松灯", "url": "https://www.bilibili.com/video/BV1BPgq6hEey/"},     {"title": "𝑩𝒊𝒍𝒍𝒊𝒆 𝑱𝒆𝒂𝒏 —— 千早爱音", "url": "https://www.bilibili.com/video/BV1dQKY6ME3t/"}

{"title": "𝑺𝑻𝑨𝑹𝑩𝑶𝒀 高松灯", "url": "https://www.bilibili.com/video/BV1BPgq6hEey/"},

{"title": "𝑩𝒊𝒍𝒍𝒊𝒆 𝑱𝒆𝒂𝒏 —— 千早爱音", "url": "https://www.bilibili.com/video/BV1dQKY6ME3t/"}

#!/usr/bin/env python3

- [𝑺𝑻𝑨𝑹𝑩𝑶𝒀 高松灯](https://www.bilibili.com/video/BV1BPgq6hEey/)

- [𝑩𝒊𝒍𝒍𝒊𝒆 𝑱𝒆𝒂𝒏 —— 千早爱音](https://www.bilibili.com/video/BV1dQKY6ME3t/)

此模板本身不调用模型。参考成品的歌声音色来自少女乐队 DDSP 模型，感谢模型作者**路边墨缇斯**；[原分享/教程入口](https://www.bilibili.com/video/BV1JU2LBZEt5/)，详细模型、声码器与许可见[翻唱模板来源](../../singing-cover/girls-band-ddsp/sources.md)。不能把这份致谢当作原曲、录音、角色、声优或字体授权。

本仓库原创渲染代码依仓库 MIT 许可；[Pillow](https://python-pillow.github.io/) 与 [FFmpeg](https://ffmpeg.org/) 依各自许可。用户图片、字体、歌词、原曲和录音不转授 MIT。发布生成内容前单独核对使用权。

Executable file

[Credential-related source omitted]

示例：[竖屏 ComfyUI 展示动画（B站）](https://www.bilibili.com/video/BV1Qqto6HE8m)。

- [shenrui-comfyui-toolkit](https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit)：本仓库其它模板使用的固定版本工作流；本模板不调用模型推理

- [Shenrui-Ma/video-regen-assets](https://huggingface.co/datasets/Shenrui-Ma/video-regen-assets)：公开素材库；产品图与字体由用户自备

Executable file

#!/usr/bin/env python3

Executable file

Executable file

Executable file

Executable file

- **RVC 官方实现**（[RVC-Project，固定 commit 4338f12](https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI/tree/4338f12c3c28c80b3ac015e2d0df66c41592746d)）：代码 MIT，不覆盖全部权重与录音。

- **HF HuBERT / RMVPE / v2 预训练权重与索引**（[VoiceConversionWebUI 资产入口](https://huggingface.co/lj1995/VoiceConversionWebUI/tree/main)）：`main` 可变，历史下载 revision 未恢复；逐项许可需自行核对，期望字节见 [model-catalog/dependencies.json](model-catalog/dependencies.json)。

- **pymss 与分离模型**：pymss 代码 MIT；其部署镜像卡片的许可不自动证明原作者授予再许可。只复用分离/声部听审方法。 - **角色、录音与声音权益**：历史来源为[【明日方舟中文语音】莫斯提马（CV. 若舞）](https://www.bilibili.com/video/BV17a411d7ex/)的转载合集，角色与游戏权利见[明日方舟官方站](https://ak.hypergryph.com/)。**不把若舞列为本合成演唱的参与者或背书人**；合成歌

- **角色、录音与声音权益**：历史来源为[【明日方舟中文语音】莫斯提马（CV. 若舞）](https://www.bilibili.com/video/BV17a411d7ex/)的转载合集，角色与游戏权利见[明日方舟官方站](https://ak.hypergryph.com/)。**不把若舞列为本合成演唱的参与者或背书人**；合成歌曲应披露 AI 音色转换。

- **工具**：[yt-dlp](https://github.com/yt-dlp/yt-dlp)、[FFmpeg](https://ffmpeg.org/) 依各自许可。

[BV17a411d7ex](https://www.bilibili.com/video/BV17a411d7ex/) 是转载/整理语音入口，**不是官方账号或本次 RVC 翻唱成品**。本模板不附权重、索引、录音、ASR 全文，不用其他角色成品冒充本次成果。历史训练和成品证据的限制见[来源与致谢](sources.md)。

- [shenrui-comfyui-toolkit](https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit)：本仓库其它模板使用的固定版本工作流；本模板是音频链路，不调用它做推理

- [Shenrui-Ma/video-regen-assets](https://huggingface.co/datasets/Shenrui-Ma/video-regen-assets)：公开素材库；本模板不附带权重、索引或录音

"schema_version": 1,   "compatibility_profile": "rvc-4338f12-v2-48k-f0-hf-hubert",   "code_url": "https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI/tree/4338f12c3c28c80b3ac015e2d0df66c41592746d",

"compatibility_profile": "rvc-4338f12-v2-48k-f0-hf-hubert",   "code_url": "https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI/tree/4338f12c3c28c80b3ac015e2d0df66c41592746d",   "asset_source_url": "https://huggingface.co/l

"code_url": "https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI/tree/4338f12c3c28c80b3ac015e2d0df66c41592746d",

"asset_source_url": "https://huggingface.co/lj1995/VoiceConversionWebUI/tree/main",   "asset_download_revision": null,   "hash_basis": "Imported from v1.0 source-package read-only audit records; not newly downloaded or hashed by this templa

"language": "zh-CN",     "actor": "若舞",     "source_url": "https://www.bilibili.com/video/BV17a411d7ex/",

"actor": "若舞",     "source_url": "https://www.bilibili.com/video/BV17a411d7ex/",     "source_kind": "reposted-voice-compilation-not-official-account",

"source_url": "https://www.bilibili.com/video/BV17a411d7ex/",

test ! -e "$JOB" mkdir -p "$JOB" git clone https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI.git "$RVC_APP"

mkdir -p "$JOB" git clone https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI.git "$RVC_APP" git -C "$RVC_APP" checkout --detach 4338f12c3c28c80b3ac015e2d0df66c41592746d

git clone https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI.git "$RVC_APP"

cd "$RVC_APP" python3.12 -m venv .venv "$RVC_PY" -m pip install torch==2.7.1+cu118 torchaudio==2.7.1+cu118 --index-url https://download.pytorch.org/whl/cu118

python3.12 -m venv .venv "$RVC_PY" -m pip install torch==2.7.1+cu118 torchaudio==2.7.1+cu118 --index-url https://download.pytorch.org/whl/cu118 "$RVC_PY" -m pip install -r requirments_cu118_py312.txt

"$RVC_PY" -m pip install torch==2.7.1+cu118 torchaudio==2.7.1+cu118 --index-url https://download.pytorch.org/whl/cu118

上游公开资产入口为 [lj1995/VoiceConversionWebUI](https://huggingface.co/lj1995/VoiceConversionWebUI/tree/main)，mute 来自 `mute.zip`，解压前检查成员路径并放入新暂存目录，再核对上述布局，防止多套一层目录或覆盖旧资产。不要下载/反序列化来源不明的 pickle、checkpoint；SHA 一致只证明字节，不证明文件本身可信。

1. **官方入口**：搜索“作品 角色 语言 语音/台词/voice lines 配音演员”，查发行方角色页、官方语音资料/媒体库与官方发布账号。记录域名、页面标题、作者/账号、语言和实际音频链接；需要登录或没有音频时如实记下。 2. **可靠角色资料站**：从官方角色身份定位对应语音记录。明日方舟可查 [PRTS 莫斯提马](https://prts.wiki/w/莫斯提马)及[语音记录](https://prts.wiki/w/莫斯提马/语音记录)、[wiki.gg](

2. **可靠角色资料站**：从官方角色身份定位对应语音记录。明日方舟可查 [PRTS 莫斯提马](https://prts.wiki/w/莫斯提马)及[语音记录](https://prts.wiki/w/莫斯提马/语音记录)、[wiki.gg](https://arknights.wiki.gg/wiki/Mostima/Audio)。这些是社区资料站，不是发行方官网。读取页面实际 `<audio>`/下载链接和语言标签；不能凭文件名规律拼造几十条 URL，也不能将中文台词

[discover.py](../scripts/discover.py) 调用 **yt-dlp 内建搜索**，只读取元数据，不下载或训练。它不是通用全网搜索引擎，也不会自动认证官方身份；官网/资料站检索仍按上节由 Agent 完成。脚本只需 Python 标准库，联网检索另需已安装的 [yt-dlp](https://github.com/yt-dlp/yt-dlp)。没有安装时按上游说明在独立工具环境准备，例如 `uv tool install yt-dlp`，不要改 

原中文来源为 [BV17a411d7ex](https://www.bilibili.com/video/BV17a411d7ex/)《【明日方舟中文语音】莫斯提马【CV. 若舞】》，简介指向 PRTS。它是转载整理入口，不是官方训练许可。历史先遭遇资料站访问失败，再通过 Bilibili 元数据搜索取得此合集；本模板将这个解决过程泛化，而非只硬编码该 BV。后续换角色必须重新找其本人的原声。

官方 [`infer/vc/modules.py`](https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI/blob/4338f12c3c28c80b3ac015e2d0df66c41592746d/infer/vc/modules.py) 的签名为：

[Credential-related source omitted]

[Credential-related source omitted]

[Credential-related source omitted]

[Credential-related source omitted]

[Credential-related source omitted]

[Credential-related source omitted]

[Credential-related source omitted]

[Credential-related source omitted]

[Credential-related source omitted]

#!/usr/bin/env python3

#!/usr/bin/env python3

raise ValueError('Bilibili p must be one positive integer')             suffix = '?p=' + parts[0] if parts else ''             return 'bilibili', 'https://www.bilibili.com/video/' + match.group(1) + '/' + suffix

suffix = '?p=' + parts[0] if parts else ''             return 'bilibili', 'https://www.bilibili.com/video/' + match.group(1) + '/' + suffix     if host in ('www.youtube.com', 'youtube.com', 'm.youtube.com') and parsed.path == '/watch':

return 'bilibili', 'https://www.bilibili.com/video/' + match.group(1) + '/' + suffix

ids = parse_qs(parsed.query).get('v', [])         if len(ids) == 1 and re.fullmatch(r'[A-Za-z0-9_-]{11}', ids[0]):             return 'youtube', 'https://www.youtube.com/watch?v=' + ids[0]

if len(ids) == 1 and re.fullmatch(r'[A-Za-z0-9_-]{11}', ids[0]):             return 'youtube', 'https://www.youtube.com/watch?v=' + ids[0]     if host == 'youtu.be' and re.fullmatch(r'/[A-Za-z0-9_-]{11}', parsed.path):

return 'youtube', 'https://www.youtube.com/watch?v=' + ids[0]

if host == 'youtu.be' and re.fullmatch(r'/[A-Za-z0-9_-]{11}', parsed.path):         return 'youtube', 'https://www.youtube.com/watch?v=' + parsed.path[1:]     raise ValueError('Use a Bilibili video or YouTube watch page, not a media/downloa

return 'youtube', 'https://www.youtube.com/watch?v=' + parsed.path[1:]

url = raw.get('webpage_url') or raw.get('url')     if provider == 'youtube' and isinstance(url, str) and re.fullmatch(r'[A-Za-z0-9_-]{11}', url):         url = 'https://www.youtube.com/watch?v=' + url

if provider == 'youtube' and isinstance(url, str) and re.fullmatch(r'[A-Za-z0-9_-]{11}', url):         url = 'https://www.youtube.com/watch?v=' + url     site, url = canonical_url(url)

url = 'https://www.youtube.com/watch?v=' + url

#!/usr/bin/env python3

| 来源 | 用途与边界 | | --- | --- | | [RVCProject / RVC-Project](https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI/tree/4338f12c3c28c80b3ac015e2d0df66c41592746d) | 感谢官方 RVC 实现；本模板固定此 commit，代码许可 MIT 不覆盖全部权重与录音 |

| --- | --- | | [RVCProject / RVC-Project](https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI/tree/4338f12c3c28c80b3ac015e2d0df66c41592746d) | 感谢官方 RVC 实现；本模板固定此 commit，代码许可 MIT 不覆盖全部权重与录音 | | [VoiceConversionWebUI 资产入口](

| [RVCProject / RVC-Project](https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI/tree/4338f12c3c28c80b3ac015e2d0df66c41592746d) | 感谢官方 RVC 实现；本模板固定此 commit，代码许可 MIT 不覆盖全部权重与录音 |

| [VoiceConversionWebUI 资产入口](https://huggingface.co/lj1995/VoiceConversionWebUI/tree/main) | HF HuBERT、RMVPE、v2 预训练与 mute 来源索引；main 可变，原下载 revision 未恢复，需独立查许可和固定 revision |

| [BV17a411d7ex](https://www.bilibili.com/video/BV17a411d7ex/) | 历史中文源：【明日方舟中文语音】莫斯提马【CV. 若舞】；转载/整理合集，不是官方账号，不是 RVC 成品 |

| [PRTS 莫斯提马](https://prts.wiki/w/莫斯提马)、[语音记录](https://prts.wiki/w/莫斯提马/语音记录) | 可信度需按具体页面核对的资料站；转载页简介所指来源，不是训练许可。历史直取曾遭 403 |

| [明日方舟](https://ak.hypergryph.com/) | 角色与游戏权利主体的官方入口；感谢原角色中文配音若舞及游戏权利主体（鹰角网络等，具体权利以各项声明为准） |

| [yt-dlp](https://github.com/yt-dlp/yt-dlp)、[FFmpeg](https://ffmpeg.org/)、[ffprobe](https://ffmpeg.org/ffprobe.html) | 来源元数据、合法取得材料后的解码和技术检查工具，工具可用不代表获得素材使用权 |

Executable file

spec.loader.exec_module(m) IDENTITY = {'character': '莫斯提马', 'series': '明日方舟', 'language': 'zh-CN', 'actor': '若舞', 'aliases': ['Mostima']} ITEM = {'title': '【明日方舟中文语音】莫斯提马【CV. 若舞】', 'url': 'https://www.bilibili.com/video/BV17a411d7ex',

IDENTITY = {'character': '莫斯提马', 'series': '明日方舟', 'language': 'zh-CN', 'actor': '若舞', 'aliases': ['Mostima']} ITEM = {'title': '【明日方舟中文语音】莫斯提马【CV. 若舞】', 'url': 'https://www.bilibili.com/video/BV17a411d7ex',         'duration': 246.677, 'up

ITEM = {'title': '【明日方舟中文语音】莫斯提马【CV. 若舞】', 'url': 'https://www.bilibili.com/video/BV17a411d7ex',

def test_url_credentials_media_and_private_hosts_rejected(self):         for url in ['file:///tmp/voice.wav', 'http://localhost/voice', 'https://user:pass@www.bilibili.com/video/BV17a411d7ex',

for url in ['file:///tmp/voice.wav', 'http://localhost/voice', 'https://user:pass@www.bilibili.com/video/BV17a411d7ex',

'https://www.bilibili.com.evil.invalid/video/BV17a411d7ex',

[Credential-related source omitted]

[Credential-related source omitted]

self.assertEqual(m.canonical_url('https://youtu.be/abcdefghijk?t=4')[1], 'https://www.youtube.com/watch?v=abcdefghijk')

[Credential-related source omitted]

if 'bilisearch' in command[-1]:                 return subprocess.CompletedProcess(command, 1, '', '403 Cookie=secret')             return self.response([{**ITEM, 'url': 'https://www.youtube.com/watch?v=abcdefghijk'}])

return subprocess.CompletedProcess(command, 1, '', '403 Cookie=secret')             return self.response([{**ITEM, 'url': 'https://www.youtube.com/watch?v=abcdefghijk'}])         report = m.discover(IDENTITY, ['bilibili', 'youtube'], run=ru

return self.response([{**ITEM, 'url': 'https://www.youtube.com/watch?v=abcdefghijk'}])

def test_deduplication_bounds_and_host_mismatch(self):         rows = [ITEM, {**ITEM, 'url': ITEM['url'] + '?tracking=a'},                 {**ITEM, 'url': 'https://www.youtube.com/watch?v=abcdefghijk'}]

rows = [ITEM, {**ITEM, 'url': ITEM['url'] + '?tracking=a'},                 {**ITEM, 'url': 'https://www.youtube.com/watch?v=abcdefghijk'}]         found, rejected = m.decode_candidates(json.dumps({'entries': rows}), 'bilibili', IDENTITY, 1

{**ITEM, 'url': 'https://www.youtube.com/watch?v=abcdefghijk'}]

def test_bilibili_url_only_flat_results_require_inspection_not_rejection(self):         row = {'_type': 'url', 'url': 'https://www.bilibili.com/video/av259650956', 'ie_key': 'BiliBili', 'id': '259650956'}

row = {'_type': 'url', 'url': 'https://www.bilibili.com/video/av259650956', 'ie_key': 'BiliBili', 'id': '259650956'}

Executable file

- **DDSP-SVC 6.2**（[yxlllc，固定 commit aeaba53f](https://github.com/yxlllc/DDSP-SVC/commit/aeaba53f46fbb4e8177af444c5c56660f55b1787)）：代码 MIT，不覆盖全部权重。

- **PC-NSF-HiFiGAN 2025.02**：资料保存的发行声明为**预训练权重 CC BY-NC-SA 4.0（非商业）**；使用或再分发前请阅读原包 notice。角色免署名声明不抵消声码器的署名与非商业要求。 - **HyperACE v2 / Gabox MelBand 等分离权重**：原权重许可未厘清，未知许可不等于允许使用或商用。 - **角色模型**：感谢模型作者**路边墨缇斯**，原分享入口 [BV1JU2LBZEt5](https://www.b

- **HyperACE v2 / Gabox MelBand 等分离权重**：原权重许可未厘清，未知许可不等于允许使用或商用。 - **角色模型**：感谢模型作者**路边墨缇斯**，原分享入口 [BV1JU2LBZEt5](https://www.bilibili.com/video/BV1JU2LBZEt5/)；资料记录的合规非商业使用说明**不是完整授权文本**，也不覆盖原曲、录音、声优或角色 IP。 - **歌曲、歌词、角色与声优**：归各自权利人；本模板不附带音频，

- **角色模型**：感谢模型作者**路边墨缇斯**，原分享入口 [BV1JU2LBZEt5](https://www.bilibili.com/video/BV1JU2LBZEt5/)；资料记录的合规非商业使用说明**不是完整授权文本**，也不覆盖原曲、录音、声优或角色 IP。

| 成品 | 角色 | 阅读重点 | | --- | --- | --- | | [𝑺𝑻𝑨𝑹𝑩𝑶𝒀 高松灯](https://www.bilibili.com/video/BV1BPgq6hEey/) | 高松灯 / `tomori-30000` | 原调重推、和声阈值、四尾音修复与手动混音版本分离 |

| --- | --- | --- | | [𝑺𝑻𝑨𝑹𝑩𝑶𝒀 高松灯](https://www.bilibili.com/video/BV1BPgq6hEey/) | 高松灯 / `tomori-30000` | 原调重推、和声阈值、四尾音修复与手动混音版本分离 | | [𝑩𝒊𝒍𝒍𝒊𝒆 𝑱𝒆𝒂𝒏 —— 千早爱音](https://www.bilibili.com/video/BV1dQKY6ME3t/) | 千早爱音 / `anon-12000` | 声部分配、错误音节诊断与

| [𝑺𝑻𝑨𝑹𝑩𝑶𝒀 高松灯](https://www.bilibili.com/video/BV1BPgq6hEey/) | 高松灯 / `tomori-30000` | 原调重推、和声阈值、四尾音修复与手动混音版本分离 |

| [𝑩𝒊𝒍𝒍𝒊𝒆 𝑱𝒆𝒂𝒏 —— 千早爱音](https://www.bilibili.com/video/BV1dQKY6ME3t/) | 千早爱音 / `anon-12000` | 声部分配、错误音节诊断与窄范围返修 |

**感谢角色模型作者路边墨缇斯**，分享来源为 [BV1JU2LBZEt5](https://www.bilibili.com/video/BV1JU2LBZEt5/)。资料中的教程上传者未知，不将其自动等同于模型作者或上述成品 UP 主。DDSP-SVC 代码作者为 yxlllc；依赖、声码器、角色模型和歌曲权利各自独立，见[来源与许可](sources.md)。

- [shenrui-comfyui-toolkit](https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit)：本仓库其它模板使用的固定版本工作流；本模板是音频链路，不调用它做推理

- [Shenrui-Ma/video-regen-assets](https://huggingface.co/datasets/Shenrui-Ma/video-regen-assets)：公开素材库；本模板不附带权重与音频

{       "id": "contentvec-legacy-500", "stage": "ddsp", "author": "auspicious3000 / ContentVec",       "source_url": "https://github.com/auspicious3000/contentvec",

"id": "contentvec-legacy-500", "stage": "ddsp", "author": "auspicious3000 / ContentVec",       "source_url": "https://github.com/auspicious3000/contentvec",       "download_url": "https://ibm.box.com/s/z1wgl1stco8ffooyatzdwsqn2psd9lrr",

"source_url": "https://github.com/auspicious3000/contentvec",

"download_url": "https://ibm.box.com/s/z1wgl1stco8ffooyatzdwsqn2psd9lrr",

{       "id": "rmvpe-230917", "stage": "ddsp", "author": "yxlllc; original research Dream-High",       "source_url": "https://github.com/yxlllc/RMVPE/releases/tag/230917",

"id": "rmvpe-230917", "stage": "ddsp", "author": "yxlllc; original research Dream-High",       "source_url": "https://github.com/yxlllc/RMVPE/releases/tag/230917",       "license_status": "Selected weight license unclear in source evidence;

"source_url": "https://github.com/yxlllc/RMVPE/releases/tag/230917",

{       "id": "pc-nsf-hifigan-2025.02", "stage": "ddsp", "author": "OpenVPI / DiffSinger Community Vocoder Project",       "source_url": "https://github.com/openvpi/vocoders/releases/tag/pc-nsf-hifigan-44.1k-hop512-128bin-2025.02",

"id": "pc-nsf-hifigan-2025.02", "stage": "ddsp", "author": "OpenVPI / DiffSinger Community Vocoder Project",       "source_url": "https://github.com/openvpi/vocoders/releases/tag/pc-nsf-hifigan-44.1k-hop512-128bin-2025.02",       "license_s

"source_url": "https://github.com/openvpi/vocoders/releases/tag/pc-nsf-hifigan-44.1k-hop512-128bin-2025.02",       "license_status": "Pretrained weights CC BY-NC-SA 4.0 per recorded release; read archive notice before use or redistribution.

{       "id": "bs_roformer_voc_hyperacev2", "stage": "separation", "author": "pcunwa / unwa",       "source_url": "https://huggingface.co/pcunwa/BS-Roformer-HyperACE",

"id": "bs_roformer_voc_hyperacev2", "stage": "separation", "author": "pcunwa / unwa",       "source_url": "https://huggingface.co/pcunwa/BS-Roformer-HyperACE",       "license_status": "Original weights unclear; mirror Apache-2.0 is not proo

"source_url": "https://huggingface.co/pcunwa/BS-Roformer-HyperACE",

"license_status": "Original weights unclear; mirror Apache-2.0 is not proof of original authorization.",       "artifacts": [         {"path": "separation/vocal/vocal_extraction/bs_roformer_voc_hyperacev2.ckpt", "bytes": 288724430, "sha256"

"artifacts": [         {"path": "separation/vocal/vocal_extraction/bs_roformer_voc_hyperacev2.ckpt", "bytes": 288724430, "sha256": "54cf516f621f2f460bf660ed137e244b8931bf7a2ce85ddceecff816dbc4d668", "source_url": "https://huggingface.co/pcu

{"path": "separation/vocal/vocal_extraction/bs_roformer_voc_hyperacev2.ckpt", "bytes": 288724430, "sha256": "54cf516f621f2f460bf660ed137e244b8931bf7a2ce85ddceecff816dbc4d668", "source_url": "https://huggingface.co/pcunwa/BS-Roformer-HyperAC

{"path": "separation/vocal/vocal_extraction/bs_roformer_voc_hyperacev2.yaml", "bytes": null, "sha256": "18c9edecb4168e92b3aaba6b3b9c22448b3a4ebc445becc4b4ae808e57d29f7f", "source_url": "https://huggingface.co/baicai1145/pymss/resolve/f02318

{       "id": "bs_karaoke_gabox_IS", "stage": "separation", "author": "GaboxR67",       "source_url": "https://huggingface.co/GaboxR67/MelBandRoformers",

"id": "bs_karaoke_gabox_IS", "stage": "separation", "author": "GaboxR67",       "source_url": "https://huggingface.co/GaboxR67/MelBandRoformers",       "license_status": "Original weights unclear; mirror Apache-2.0 is not proof of original 

"source_url": "https://huggingface.co/GaboxR67/MelBandRoformers",

"license_status": "Original weights unclear; mirror Apache-2.0 is not proof of original authorization.",       "artifacts": [         {"path": "separation/karaoke/bs_karaoke_gabox_IS.ckpt", "bytes": 204483448, "sha256": "db8357825398d423103

"artifacts": [         {"path": "separation/karaoke/bs_karaoke_gabox_IS.ckpt", "bytes": 204483448, "sha256": "db8357825398d4231031ad1ab4aa12a94bcaad8d67e8ce5e4b3c5b48fdee1d4f", "source_url": "https://huggingface.co/GaboxR67/MelBandRoformers

{"path": "separation/karaoke/bs_karaoke_gabox_IS.ckpt", "bytes": 204483448, "sha256": "db8357825398d4231031ad1ab4aa12a94bcaad8d67e8ce5e4b3c5b48fdee1d4f", "source_url": "https://huggingface.co/GaboxR67/MelBandRoformers/resolve/7860d31d8cc1c7

{"path": "separation/karaoke/bs_karaoke_gabox_IS.yaml", "bytes": 2004, "sha256": "8360384097cc7a0b229b2918f70270bf7f5e64630d933ff359ceb6eaadf40fd9", "source_url": "https://huggingface.co/baicai1145/pymss/resolve/f02318ef92c43472777ed7aeeded

{       "id": "Lead_VocalDereverb", "stage": "separation", "author": "GaboxR67",       "source_url": "https://huggingface.co/GaboxR67/MelBandRoformers",

"id": "Lead_VocalDereverb", "stage": "separation", "author": "GaboxR67",       "source_url": "https://huggingface.co/GaboxR67/MelBandRoformers",       "license_status": "Original weights unclear; mirror Apache-2.0 is not proof of original a

"source_url": "https://huggingface.co/GaboxR67/MelBandRoformers",

"known_bad_copy": {"bytes": 255995904, "sha256": "e227c8ba976052c39d2206174ed51092538b70a5086e0732cf1fd57d80af1339", "action": "Reject this historical corrupt or incomplete copy; do not alter expected hashes."},       "artifacts": [        

"artifacts": [         {"path": "separation/reverb_echo_control/dereverb/Lead_VocalDereverb.ckpt", "bytes": 913031195, "sha256": "097c13b298414905651867ef315d881bbf2604a5c04be28820ed5946338b3eb4", "source_url": "https://huggingface.co/Gabox

{"path": "separation/reverb_echo_control/dereverb/Lead_VocalDereverb.ckpt", "bytes": 913031195, "sha256": "097c13b298414905651867ef315d881bbf2604a5c04be28820ed5946338b3eb4", "source_url": "https://huggingface.co/GaboxR67/MelBandRoformers/re

{"path": "separation/reverb_echo_control/dereverb/Lead_VocalDereverb.yaml", "bytes": 1315, "sha256": "075e90763f8a0048c13085b7ea0141875a73f19b8ecd1223599d06ea14661cca", "source_url": "https://huggingface.co/baicai1145/pymss/resolve/f02318ef

"author_evidence": "source-package-search-index-excerpt; tutorial uploader unknown",   "tutorial_uploader": null,   "source_url": "https://www.bilibili.com/video/BV1JU2LBZEt5/",

"tutorial_uploader": null,   "source_url": "https://www.bilibili.com/video/BV1JU2LBZEt5/",   "usage_notice": "Source package records custom non-commercial terms; review original notice and all other rights before use. No authorization is in

"source_url": "https://www.bilibili.com/video/BV1JU2LBZEt5/",

},   "examples": [     {"title": "𝑺𝑻𝑨𝑹𝑩𝑶𝒀 高松灯", "url": "https://www.bilibili.com/video/BV1BPgq6hEey/", "voice_id": "tomori-30000"},

"examples": [     {"title": "𝑺𝑻𝑨𝑹𝑩𝑶𝒀 高松灯", "url": "https://www.bilibili.com/video/BV1BPgq6hEey/", "voice_id": "tomori-30000"},     {"title": "𝑩𝒊𝒍𝒍𝒊𝒆 𝑱𝒆𝒂𝒏 —— 千早爱音", "url": "https://www.bilibili.com/video/BV1dQKY6ME3t/", "voice_id": "anon-120

{"title": "𝑺𝑻𝑨𝑹𝑩𝑶𝒀 高松灯", "url": "https://www.bilibili.com/video/BV1BPgq6hEey/", "voice_id": "tomori-30000"},

{"title": "𝑩𝒊𝒍𝒍𝒊𝒆 𝑱𝒆𝒂𝒏 —— 千早爱音", "url": "https://www.bilibili.com/video/BV1dQKY6ME3t/", "voice_id": "anon-12000"}

#!/usr/bin/env python3

**明确感谢角色模型作者路边墨缇斯。** 原分享/教程来源：[BV1JU2LBZEt5](https://www.bilibili.com/video/BV1JU2LBZEt5/)。作者身份及非商业使用说明来自资料包保存的搜索索引摘录；当时 Bilibili API 返回 412，上传者未知。因此不把模型作者自动等同于教程上传者，也不推断成品 UP 主身份。

- [𝑺𝑻𝑨𝑹𝑩𝑶𝒀 高松灯](https://www.bilibili.com/video/BV1BPgq6hEey/)

- [𝑩𝒊𝒍𝒍𝒊𝒆 𝑱𝒆𝒂𝒏 —— 千早爱音](https://www.bilibili.com/video/BV1dQKY6ME3t/)

| 组件 | 作者 / 固定来源 | 资料中的许可边界 | | --- | --- | --- | | DDSP-SVC 6.2 | [yxlllc / aeaba53f](https://github.com/yxlllc/DDSP-SVC/commit/aeaba53f46fbb4e8177af444c5c56660f55b1787) | 代码 MIT，不覆盖全部权重 |

| --- | --- | --- | | DDSP-SVC 6.2 | [yxlllc / aeaba53f](https://github.com/yxlllc/DDSP-SVC/commit/aeaba53f46fbb4e8177af444c5c56660f55b1787) | 代码 MIT，不覆盖全部权重 | | pymss 2.0.14 | [pymss-project / dfe1def1](https://github.com/pymss-project/pym

| DDSP-SVC 6.2 | [yxlllc / aeaba53f](https://github.com/yxlllc/DDSP-SVC/commit/aeaba53f46fbb4e8177af444c5c56660f55b1787) | 代码 MIT，不覆盖全部权重 |

| pymss 2.0.14 | [pymss-project / dfe1def1](https://github.com/pymss-project/pymss/tree/dfe1def1ee9b880b5380b83e5e5b920eed631774) | 代码 MIT；core 0.1.4 与环境其余依赖逐项保留许可 |

| ContentVec legacy 500 | [auspicious3000 / ContentVec](https://github.com/auspicious3000/contentvec) | 代码 MIT，Box 权重独立 notice 未齐全；500 不是训练步数 | | RMVPE 230917 | [yxlllc 发行](https://github.com/yxlllc/RMVPE/releases/tag/230917)；原论文实现 Dream-Hi

| RMVPE 230917 | [yxlllc 发行](https://github.com/yxlllc/RMVPE/releases/tag/230917)；原论文实现 Dream-High | 所用权重许可不清楚，不能套用 RVC 或另一个实现的许可 | | PC-NSF-HiFiGAN 2025.02 | [OpenVPI / DiffSinger Community Vocoder Project](https://github.com/openvpi/vocod

| PC-NSF-HiFiGAN 2025.02 | [OpenVPI / DiffSinger Community Vocoder Project](https://github.com/openvpi/vocoders/releases/tag/pc-nsf-hifigan-44.1k-hop512-128bin-2025.02) | 资料保存的发行声明为预训练权重 CC BY-NC-SA 4.0；使用/再分发前阅读原包 notice |

| HyperACE v2 | [pcunwa / unwa](https://huggingface.co/pcunwa/BS-Roformer-HyperACE) | 原权重许可未厘清 |

| Gabox karaoke / Lead_VocalDereverb | [GaboxR67](https://huggingface.co/GaboxR67/MelBandRoformers) | 原权重许可未厘清 |

分离配对配置引用 [pymss 部署镜像](https://huggingface.co/baicai1145/pymss/tree/f02318ef92c43472777ed7aeeded3d2b29a68b58)。镜像卡的 Apache-2.0 不自动证明原作者授予再许可；pymss 的 MIT 也不覆盖它们。哈希及固定下载入口见 [dependencies.json](model-catalog/dependencies.json)。

Executable file