EvoLink Image (All Models)

Most of this pack gives you one node per model - GPT Image 2, Nano Banana 2, Seedream 5.0, and so on. EvoLink_Image is the opposite philosophy: one node, one dropdown, and every image model the platform offers behind it, including the beta budget channels and older models that never got their own node. It's the "I'm not sure which model yet" node, and it's also the one to check first when a new model launches - EvoLink adds it to this list without a plugin update.

The catch is the same one that makes the per-model nodes worth having: a one-size node can't expose every model's specific knobs, so model-specific parameters go through an extra_params JSON field instead of a tidy widget.

The inputs that matter

• prompt (required) - your description or edit instruction; Chinese works.
• model (required) - the dropdown. It starts with the headline models (gpt-image-2, grok-imagine-image-2.0, nano-banana-2, nano-banana-pro, the Seedream and Qwen family...) and continues past the visible list into the beta channels and older tiers. Try them.
• size - a free-text field, so auto, a ratio like 1:1, or explicit pixels like 1024x1024. What's valid depends on the selected model - check the API docs if a value gets rejected.
• image1–image4 - reference image sockets. Connect Load Image to image1 for image-to-image; image_urls_text accepts one public URL per line and mixes with the sockets.
• extra_params - the escape hatch. A JSON object that gets merged verbatim into the request body. Need {"quality":"2K"} on a model that doesn't expose it as a widget? This is how.

Then api_key and timeout_seconds (default 300).

How it works

The node posts to POST /v1/images/generations with the model you picked, uploads any local reference images through EvoLink's file channel (auto-deleted after 72h, no third-party host), and polls until the image is ready. The image output feeds Preview or Save Image directly. You also get task_id, status_info, result_urls (24h expiry), and response_json. If extra_params isn't valid JSON, the pack refuses the job with a clear error rather than sending garbage upstream.

Installing

No GPU, no model files, only requests:

cd ComfyUI/custom_nodes
git clone https://github.com/deeplearning-goethe/comfyui-evolink.git

then fully restart ComfyUI, or install via ComfyUI Manager by searching EvoLink.

Troubleshooting

401 key, 402 balance, 403 model access, 429 rate limit - and with a multi-model node, 403 deserves extra attention, because your account might have access to one model in the list but not another. Also watch size: a value that works on Nano Banana can be rejected by Qwen. When in doubt, use auto. One more honest note: the README's roadmap text lags behind what the pack actually ships, so don't be surprised if a model listed as "coming soon" is already sitting in this dropdown.