comfyui-byokey
ComfyUI nodes enabling direct API key usage with custom proxy support for OpenAI, Claude, Gemini, and video/image generation services. (Description by CC)
Nodes (37)
Bria's enterprise-grade text-to-image, dropped into your workflow
ByteDance's Seed/Doubao models through the Ark API
Claude with your own key, plus vision
ElevenLabs sound effects from a text prompt — foley without the foley
ElevenLabs TTS in ComfyUI — the voice bar, without leaving the graph
Flux Pro 1.1 Ultra via the BFL API, no Comfy credits required
Vision-capable text gen with your own key
Gemini's image model with your own key
XAI's Grok image model in your graph — the one with the least filtering
Grok's video generator as a ComfyUI node — text-to-video or image-to-video
Ideogram V3's API in ComfyUI — for when text-in-image actually matters
GPT Image and DALL·E 3 in ComfyUI without Comfy's credit meter
Kling image-to-video — animate a still, optionally end on a different frame
Kling text-to-video — pick your mode/duration/model combo, go
One LLM node that talks to basically every chat API (and relays)
LTX-2 video via API — Lightricks' fast frontier, at up to 4K with audio
Luma image-to-video — animate a still with Ray 2 and your own key
Luma Dream Machine text-to-video — Ray 2 with your own key
Magnific AI upscaling — the 16x creative upscale, now a node
MiniMax (Hailuo) image-to-video — animate a still with your own key
Your own key into MiniMax's Hailuo video API, no Comfy credit meter
OpenRouter as your in-graph LLM
PixVerse t2v with your own key — the fast, cheap end of video APIs
The text-in-image specialist, 89 substyles deep
Reve create, the new closed-model darling, with your own key
Runway Gen4 i2v, own key, no credit meter
Runway makes images now, and you can drive Gen4 with your own key
Seedance 2.0 with first-and-last-frame control, straight from ByteDance
Seedance 2.0 text-to-video on your own ByteDance key
Stability's API with your own key — including the one model you can't run
Veo 2 and 3 on your own Google key — the video quality ceiling, in your graph
Point it at any submit-then-poll API
Vidu i2v with your own key — one frame in, five seconds of motion out
9, nine-sixteenths, or square, take your pick
Wan i2v the easy way — you're renting Alibaba's GPU, not downloading weights
Wan t2v from a prompt — Alibaba's open model, hosted, on your key
Cloud upscaling with AuraSR — your image in, 8x out, GPU not included
ComfyUI BYOKey — Bring Your Own Key
直接用自己的 API Key 调用 AI 服务，支持自定义中转站地址，输入/输出类型与官方节点 100% 兼容。
ComfyUI 官方 API 节点强制所有请求走 api.comfy.org 代理：用户无法使用自己的 API Key，也无法指定第三方中转站。
本项目提供 37 个轻量替代节点，覆盖所有主流官方 API 提供商。每个节点都有 api_key 和 base_url 输入框：
- 填自己的 Key，直连官方 API
- 改
base_url，走第三方中转站（对国内用户尤其重要） - 输出类型与官方节点完全一致（
VIDEO/AUDIO/IMAGE/STRING），可直接替换官方工作流中的节点
与 comfy-api-liberation 的区别
| | comfy-api-liberation | BYOKey（本项目） |
|---|---|---|
| 实现方式 | Monkey-patch 官方节点的请求 | 独立替代节点 |
| 自定义 URL | ❌ 不支持（硬编码官方地址） | ✅ 每个节点都有 base_url |
| 中转站支持 | 无 | 完整支持 |
| 稳定性 | 依赖官方内部实现，升级易失效 | 独立实现，不受官方更新影响 |
输出类型对齐官方
本项目的请求/响应格式以 官方 comfy_api_nodes/apis/*.py 的 pydantic 模型 + Comfy-Org/embedded-docs 为基准（官方代理 api.comfy.org 对 body 做透明转发，所以官方模型即真实厂商 API 契约）。
| 类型 | 官方 | BYOKey |
|------|------|--------|
| 视频 | VIDEO | ✅ VIDEO（comfy_api.latest.InputImpl.VideoFromFile） |
| 音频 | AUDIO | ✅ AUDIO（{"waveform","sample_rate"}，PyAV 解码） |
| 图像 | IMAGE | ✅ IMAGE（torch tensor，BHWC） |
| 文本 | STRING | ✅ STRING |
安装
cd ComfyUI/custom_nodes
git clone https://github.com/MeteorAndy/comfyui-byokey.git
重启 ComfyUI，节点出现在 api/byokey/... 分类下。
Node List
文本 / LLM → STRING
| 节点 | 提供商 | Vision | |------|--------|--------| | BYOKey LLM (OpenAI Compatible) | OpenAI / DeepSeek / 通义千问等 | ✅ | | BYOKey Claude (Anthropic) | Anthropic | ✅ | | BYOKey Gemini (Google) | Google | ✅ | | BYOKey OpenRouter LLM | OpenRouter（200+ 模型） | ✅ | | BYOKey ByteDance Seed LLM | ByteDance / Doubao（火山方舟） | ✅ |
图像生成 → IMAGE
| 节点 | 提供商 | |------|--------| | BYOKey GPT Image (OpenAI) | OpenAI GPT Image 2 / DALL·E 3 | | BYOKey Gemini Image (Google) | Google Gemini / Nano Banana | | BYOKey Flux (BFL) | Black Forest Labs Flux Pro | | BYOKey Stability AI Image | Stability AI (SD3.5 / Ultra) | | BYOKey Grok Image (xAI) | xAI | | BYOKey Ideogram Image | Ideogram V3 | | BYOKey Recraft Image | Recraft V3 | | BYOKey Reve Image | Reve | | BYOKey Bria Image | Bria | | BYOKey Runway Text to Image | Runway Gen4 Image |
图像超分 → IMAGE
| 节点 | 提供商 | |------|--------| | BYOKey Magnific Upscale | Magnific AI | | BYOKey WaveSpeed Upscale | WaveSpeed (AuraSR) |
视频生成 → VIDEO
| 节点 | 提供商 | |------|--------| | BYOKey Seedance 2.0 Text/Image to Video | ByteDance | | BYOKey Kling Text/Image to Video | Kling | | BYOKey MiniMax Text/Image to Video | MiniMax / Hailuo | | BYOKey Runway Image to Video | Runway Gen4 Turbo | | BYOKey Luma Text/Image to Video | Luma Dream Machine | | BYOKey Google Veo Video | Google Veo 3 | | BYOKey Wan Text/Image to Video | Wan | | BYOKey PixVerse Text to Video | PixVerse | | BYOKey Vidu Text/Image to Video | Vidu | | BYOKey LTXV Text to Video | LTX Video | | BYOKey Grok Video (xAI) | xAI | | BYOKey Video Gen (Generic) | 任意 submit+poll API |
音频 → AUDIO
| 节点 | 提供商 | |------|--------| | BYOKey ElevenLabs TTS | ElevenLabs | | BYOKey ElevenLabs Sound Effects | ElevenLabs |
使用示例
用中转站访问 OpenAI
base_url 填你的中转站地址即可（适用于所有节点）：
api_key: sk-xxxx
base_url: https://your-proxy.com/v1
model: gpt-4o
用 DeepSeek（OpenAI 兼容）
api_key: sk-your-deepseek-key
base_url: https://api.deepseek.com/v1
model: deepseek-chat
用通义千问（OpenAI 兼容）
api_key: sk-xxxx
base_url: https://dashscope.aliyuncs.com/compatible-mode/v1
model: qwen-max
GPT Image 2 生图
api_key: sk-xxxx
base_url: https://api.openai.com/v1
model: gpt-image-2
size: 1536x1024
quality: high
依赖
仅需 ComfyUI 自带的依赖（torch、aiohttp、PIL、numpy、av），无额外安装。
验证与已知限制
- 请求/响应格式、模型 id、参数范围均以官方
comfy_api_nodes/源码与Comfy-Org/embedded-docs为基准对齐。 - 直连模式下，需图片输入的节点把图像作为 inline base64 data URI 发送（官方走代理时上传到 ComfyAPI 存储）；MiniMax 等 API 原生接受该格式。
- 部分厂商（如 Bria）官方仅提供 edit 类节点，本项目按其公开 text-to-image endpoint 实现。
License
MIT