ComfyUI API Toolkit
Kling AI, ElevenLabs, and Google Gemini nodes for ComfyUI
Nodes (68)
Check your ElevenLabs character balance before you burn a run
Put the secret in once, never again
Strip the background noise out of any audio in your graph
Scripted multi-voice dialogue
Get a live list of your ElevenLabs voices — clones and all
Generate a full track from a text prompt, no composer required
Type 'thunder' get thunder
Re-voice existing audio into a new voice
Speech to Text with diarization, so your transcript knows who said what
The ElevenLabs Text to Speech node, tuned like the API itself
TTS that tells you when every word lands — for lip sync and subtitles
Instant voice clone from up to eight audio samples
Save what you just heard
Invent a voice from a description — no recording required
Pick an ElevenLabs voice from a dropdown without calling the API
One node to hold your Gemini key, everywhere else just wires in
Know your prompt's token cost before you spend a cent
Turn text into vectors for search, clustering, and RAG
Edit any image with a sentence — mask optional
Gemini image generation, no weights required
Google's dedicated text-to-image, not the Nano Banana path
Gemini fills the mask, not your VRAM
What can my Gemini key actually call? This node answers instead of guessing
One dropdown for 55 Gemini models — with a guardrail that catches your own mistakes
Actual multi-turn conversation inside a ComfyUI graph
Text-to-music inside ComfyUI, if you can stand the API tax
Extend the canvas beyond its edges — Gemini outpainting that understands the scene
Turn 'a dragon' into a real prompt — Gemini rewrites your ideas before the sampler sees them
You can't actually loosen Gemini's safety
Gemini forced into your schema, not hoping for valid output
Full-featured Gemini text generation with every knob on the table
One slider that controls how hard Gemini thinks — and how big your bill gets
Gemini TTS that sounds like a human, not a robot
8 seconds of AI video with sound, the way only Google ships it
Gemini vision analysis, up to four images at once
Re-voice a video when there are multiple faces in frame
Get your image or video onto Kling's servers as an asset
Sound effects and ambience without leaving the graph
JWT auth with a debug switch
Kling avatar generation from one image
Give your Kling video a camera move without touching the prompt
The node that gets your files onto a URL, because Kling won't take them any other way
Stop guessing Kling effect names — ask the API what it actually supports
The boring-but-important node that points Kling at your saved characters and styles
Save a Kling video without turning your GPU into a RAM-based video player
Outpaint your images with Kling instead of fighting the edge yourself
Kling text-to-image straight into your ComfyUI graph
Kling's do-anything image editor, refactored as a single node
Ask Kling what's in your image, in one node
The node that makes your still image move
Make your AI video talk — audio-synced lips, no manual animation
Steal a video's motion and put it on your own character
One prompt, six shots — consistent characters across a storyboard
Pick any file in your input folder
Save any URL or file path to disk without decoding it first
Prompt to video on Kling's GPUs
Kling text-to-speech, the simple version that gets out of your way
Kling TTS when you need to actually direct the voice
Kling's cloud upscaler for images and video alike
Turn two stills into a hugging, kissing, heart-floating video
Extend a Kling clip without rerolling the whole video
The boring Kling node you didn't know you needed (no API key required)
The catch-all video node (text + images + reference clip)
Pull a soundtrack out of any clip
Dress a person in a picture of any garment
A reusable voice_id from one audio sample
23 preset voices, zero API calls
The local node that actually makes pixel art
ComfyUI API Toolkit
Four AI services. Sixty-eight nodes. One install. Zero excuses.
ComfyUI API Toolkit bundles Kling AI, ElevenLabs, Google Gemini, and a Utils collection into a single custom node pack so you can stop juggling half a dozen plugins and actually make things. Each service lives in its own module discovered automatically from the filesystem -- add a folder under services/ and it just works. If you don't have a particular SDK installed, that service quietly sits out instead of torching your entire ComfyUI startup. The rest keeps working. You're welcome.
What's Inside
ComfyUI API Toolkit v1.3.0
├── Kling AI 32 nodes (video, image, audio, effects)
├── ElevenLabs 15 nodes (TTS, voice cloning, music, transcription)
├── Google Gemini 20 nodes (entire API covered -- text, image, audio, video, embeddings)
└── Utils 1 node (pixel art resize)
----------
68 nodes total
Services
Kling AI -- 32 nodes
Kling does a lot. Embarrassingly a lot. Video from text, video from images, extend existing videos, lip sync, avatar generation, virtual try-on, motion control, effects, upscaling, audio generation, TTS, and voice cloning. If it involves moving pixels or making sound, there's probably a node for it.
Video
- Kling Text to Video
- Kling Image to Video
- Kling Video Omni
- Kling Video Extend
- Kling Lip Sync
- Kling Advanced Lip Sync
- Kling Motion Control
- Kling Avatar Generation
Image
- Kling Image Generation
- Kling Image Omni
- Kling Image Extend
- Kling Virtual Try-On
- Kling AI Multi-Shot
- Kling Image Recognize
Audio
- Kling Text to Audio
- Kling Text to Speech
- Kling TTS Advanced
- Kling Video to Audio
- Kling Voice Clone
Effects & Upscaling
- Kling Video Effects
- Kling Effect Templates
- Kling AI Upscale
Utilities
- Kling AI Authentication
- Kling Video Loader
- Kling Raw File Loader
- Kling Raw File Saver
- Kling AI Asset Upload
- Kling AI Element
- Kling Camera Control
- Kling Voice Selector
- Kling AI Cloud Uploader
- Kling Fast Video Saver
ElevenLabs -- 15 nodes
The gold standard for AI voice. Text to speech, speech to speech, voice cloning, voice design, sound effects, music generation, audio isolation, transcription, and multi-speaker dialogue. If you're still using robotic TTS from 2019, this pack is an intervention.
- ElevenLabs - API Key
- ElevenLabs - Voice Selector
- ElevenLabs - Fetch Voices
- ElevenLabs - Voice Clone
- ElevenLabs - Voice Design
- ElevenLabs - Voice Create
- ElevenLabs - Text to Speech
- ElevenLabs - TTS with Timestamps
- ElevenLabs - Speech to Speech
- ElevenLabs - Sound Effects
- ElevenLabs - Audio Isolation
- ElevenLabs - Speech to Text
- ElevenLabs - Text to Dialogue
- ElevenLabs - Music Generation
- ElevenLabs - Account Info
Google Gemini -- 20 nodes
The Gemini service went from "a few image nodes" to covering the entire Gemini API. That means text, vision, image generation via both Gemini and Imagen, image editing, inpainting, outpainting, TTS with 30+ voices, music generation via Lyria, video generation via Veo, and embeddings. If Google offers it through their API, there's a node for it now.
Config -- 6 nodes
- Gemini API Key
- Gemini Model Selector
- Gemini Safety Settings
- Gemini Thinking Config
- Gemini List Available Models -- queries your API key for what's actually accessible to your account
- Gemini Token Counter
Text -- 4 nodes
- Gemini Text Generation -- 33 model options across Gemini 3 previews, 2.5, 2.0, and the full Gemma family
- Gemini Prompt Refiner
- Gemini Multi-Turn Chat
- Gemini Structured Output -- JSON-schema constrained generation
Image -- 6 nodes
- Gemini Vision Analysis
- Gemini Image Generation -- Nano Banana / 2 / Pro via generate_content
- Imagen Image Generation -- Imagen 4 Ultra / Standard / Fast via generate_images
- Gemini Image Edit
- Gemini Inpaint
- Gemini Outpaint
Audio -- 2 nodes
- Gemini Text-to-Speech -- 30+ prebuilt voices across 3 TTS models
- Gemini Music Generation -- Lyria 3 Clip and Pro
Video -- 1 node
- Gemini Video Generation -- Veo 3.1, 3.0, 2.0, preview, fast, and lite variants
Embeddings -- 1 node
- Gemini Text Embeddings -- 768 to 3072 dimensions with task-type optimization
Full model coverage
| Category | Models | |----------|--------| | Gemini text / multimodal | 13 models (Gemini 3 previews, 2.5, 2.0, latest aliases) | | Gemma open models | 8 models (3-1b through 4-31b) | | Gemini image gen | 3 Nano Banana models | | Imagen | 3 Imagen 4 models (Ultra, Standard, Fast) | | TTS | 3 TTS models + 30 prebuilt voices | | Embeddings | 2 embedding models | | Video | 6 Veo models | | Music | 2 Lyria models | | Specialized | Robotics ER, Computer Use, Deep Research, TTS previews |
Utils -- 1 node
Pixel Art Resize Resizes images to pixel art with proper palette locking across frames for animations. Uses Floyd-Steinberg dithering and scipy KDTree acceleration so palette matching doesn't take all day.
Features Worth Knowing About
Auto-discovery for services
Services are discovered from the filesystem. Drop a folder under services/ and it loads. Nothing to register, nothing to hardcode.
Graceful service fallback
Don't have google-genai installed? Gemini nodes don't load. Everything else does. Same deal for Kling and ElevenLabs. Install what you need, ignore what you don't.
Shared retry logic with exponential backoff All API nodes retry on transient failures. They back off intelligently instead of hammering the API until it blocks you. Rate limits happen; the nodes handle it.
API keys are password-masked Your keys don't show up in plaintext in your workflow. Share screenshots without accidentally donating your API credits to strangers.
Environment variable fallback
Set KLING_ACCESS_KEY + KLING_ACCESS_SECRET, ELEVENLABS_API_KEY, or GEMINI_API_KEY in your environment and the auth nodes will pick them up automatically. Good for automation, good for not hardcoding secrets.
IS_CHANGED on all API nodes ComfyUI caches node outputs. API nodes that cache are useless. Every API node here implements IS_CHANGED so you get fresh results every time you queue, not whatever you got the last time.
Tooltips on every parameter Hover over any input and there's a tooltip explaining what it does. Shockingly underrated feature.
Output directory auto-creation If your output path doesn't exist, it gets created. No silent failures because someone forgot to mkdir.
Cloud uploader auto-fallback The cloud uploader tries catbox first, falls back to tmpfiles if it fails. Assets get uploaded regardless.
Installation
ComfyUI Manager (easiest) Search for "API Toolkit" and install. Done.
Registry
comfy node registry-install comfyui-api-toolkit
Manual
cd ComfyUI/custom_nodes
git clone https://github.com/IxMxAMAR/ComfyUI-API-Toolkit
cd ComfyUI-API-Toolkit
pip install -r requirements.txt
Dependencies
Install all services at once:
pip install requests PyJWT soundfile scipy "google-genai>=0.8.0"
Or pick what you actually use:
# Kling only
pip install requests PyJWT
# ElevenLabs only
pip install requests soundfile scipy
# Gemini only
pip install "google-genai>=0.8.0"
The pack uses optional dependencies, so you're not forced to install Google's SDK just because you want to run some TTS.
Also Available Separately
Prefer to install just one service? Each service is published as its own standalone package:
| Package | Service | |---------|---------| | ComfyUI-Kling-Direct | Kling AI | | ComfyUI-ElevenLabs-Pro | ElevenLabs | | ComfyUI-NanoBanana2 | Google Gemini |
Same nodes, same quality, just smaller installs if you only need one thing.
Requirements
- ComfyUI (recent enough to support custom nodes, which if you're reading this you probably already have)
- Python 3.10+
- API keys for whichever services you want to use
License
MIT. Use it, fork it, build on it.
Made by IxMxAMAR