EvoLink Qwen Image Edit Plus (Official)

This is the node that finally makes the "just tell it what to change" workflow real in ComfyUI. EvoLink Qwen Image Edit Plus (Official) is an instruction-driven image editor from Alibaba's Qwen line: you hand it a picture and write the edit - "change the sign text to 'OPEN'," "re-style this as watercolor" - and it returns an edited image. No masking, no inpainting masks, no ControlNet. The instruction is the tool.

The Qwen Edit family has a serious reputation in this space: it's the lineage the open community adopted as the default editor, and the "plus" here is the higher-end closed API variant. You can't download these weights, so this node is the access path.

The one input you must not forget

Unlike the other image nodes in the pack, this one is editing-first, and image1 is required - the node won't run without it. Wire a Load Image into image1 and you're live. image2 and image3 are optional extra references, and there's an image_urls_text box for public URLs. Your local images upload through EvoLink's channel and are auto-deleted after 72 hours, same as everywhere else in this pack.

The rest of the surface:

• prompt (required) - the edit instruction. "把招牌文字改成…" or its English equivalent; the tooltip's own example is changing sign text.
• size - leave blank to follow the input image; or give a ratio/pixels to force output dimensions.
• n - 1 to 6 edited variants per run.
• prompt_extend - on by default here (this node defaults it differently than the generation nodes), a smart expander for terse instructions.
• negative_prompt, watermark, seed - the usual crew. Fixed seed for reproducible edits.
• api_key, timeout_seconds - standard.

Outputs: image (downloaded tensor), task_id, status_info, result_urls (24-hour expiry), response_json.

Install and key

Standard EvoLink install. ComfyUI Manager → search EvoLink → install → full restart (close the console, don't just refresh). Manual:

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

Only requests for a dependency; no models, no GPU. Key from evolink.ai/dashboard/keys (sk-), pasted once into api_key - saved locally, blank after, never in a shared workflow.

Where it shines and where it trips

It shines at localized edits: text swaps, style changes, object tweaks, background moves - the stuff that used to mean masking plus an inpaint pass. It trips when you ask for surgical changes and the model decides to "helpfully" touch up the whole image; that's the nature of instruction editing, and the fix is more specific prompts, not more knobs. Watch out for the recurring EvoLink gotcha (edit finishes, nothing visible → connect Save Image and re-run; cached, no double charge), and remember the cloud deal: per-call cost, your images leave the machine. For changing a real photo, that's a privacy consideration to actually weigh rather than shrug past.