ComfyUI-Kavel
Generate and edit images through the kavel.ai API. Works without an API key on the free tier; add a key to use your account's models (Nano Banana 2, GPT Image 2, Seedream).
ComfyUI-Kavel
Two ComfyUI nodes that generate and edit images through Kavel — no API key needed to start, and no GPU, no checkpoints, no VRAM.
| Node | In | Out |
|---|---|---|
| Kavel Generate Image | prompt, aspect ratio | IMAGE, url |
| Kavel Edit Image | IMAGE + one-sentence instruction | IMAGE, url |
Drop them anywhere in a workflow: generate a base image and pipe it into an upscaler, or take any
IMAGE from your graph and edit it in plain language — "put the product on white marble, keep the
label unchanged".
Install
ComfyUI-Manager: search for Kavel and install.
Manually:
cd ComfyUI/custom_nodes
git clone https://github.com/hanshs474/ComfyUI-Kavel
Restart ComfyUI. No extra Python packages — the nodes use the standard library plus the torch, numpy and Pillow that ComfyUI already ships.
Free tier, then your account
Leave api_key empty and the nodes run on Kavel's free tier: no account, 1K output with a watermark, a few images a day per machine. Enough to wire up a workflow and see it work.
Paste a key from kavel.ai/settings/apikeys (or set
KAVEL_API_KEY before starting ComfyUI) and every run goes through your account instead: your credits,
no daily ceiling, no watermark on a paid plan, and any image model your plan includes via the model
field —
Nano Banana 2 for multi-reference edits,
GPT Image 2 for 4K output,
Qwen Image 3 for pictures that contain legible text,
Seedream 5.0 Pro for photoreal people.
Pricing lists what each plan includes.
Edits that hold
The edit node runs Nano Banana 2 Lite on the free tier. It keeps what it can see and invents what it cannot, so say what must not change as clearly as what should. The same engine powers the AI hairstyle changer and the AI outfit generator — their preset prompts are good templates. Your input image is uploaded to Kavel's CDN so the model can read it.
Notes
- A run waits 25–80 seconds in the free queue before the model starts; paid plans skip it.
- Both nodes also output the permanent CDN
url, handy for logging or passing to a web hook. - Change seed to force a fresh run on an unchanged prompt (ComfyUI caches identical inputs).
- Video: use the AI video generator on the site.
Want the same thing from code? SDKs for Python, Go, Rust, TypeScript, PHP, Ruby, Dart and .NET.
MIT