EvoLink Qwen Image 3.0 Pro (Official)

If Qwen Image 3.0 is the capable daily driver of Alibaba's image line, the Pro tier is the one where Alibaba lets the model think longer and render harder. Same family, same skill set - especially that text-in-the-frame rendering that made the Qwen Image line famous - but higher quality and, naturally, a pricier per-call meter. EvoLink Qwen Image 3.0 Pro (Official) is the closed API version dropped into ComfyUI: prompt and optional references in, finished image out, nothing running on your machine.

What it is

Text-to-image and image-to-image, with exactly the same control surface as the base 3.0 node. Don't expect new toys - expect the same toys with better output:

• prompt (required) - description, Chinese welcome. Text-in-image jobs: spell out exactly what should appear in the frame.
• size - auto, a ratio like 16:9, or pixel values.
• quality - 1K or 2K, default 1K.
• n - 1 to 6 images per call, handy for options without multiplying calls.
• negative_prompt, prompt_extend, watermark, seed - the same gang as 3.0. Fixed seed reproduces a result; prompt_extend helps short prompts.
• image1–image4 - reference images for img2img, uploaded via EvoLink's channel and deleted after 72 hours.

Outputs are the pack standard: image (downloaded tensor), task_id, status_info, result_urls (24-hour expiry), response_json. Remember the wire-up: image into Preview Image or Save Image, or the finished render stays invisible.

Install and key

Shared EvoLink install. ComfyUI Manager → search EvoLink → install → full restart (close the console window, not just refresh the tab). Manual:

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

Only requests as a dependency; no model files, no GPU. Key from evolink.ai/dashboard/keys (sk-), pasted once into api_key - it saves locally and stays blank after, so it won't ride along in a shared workflow JSON.

Pro vs. base - how to actually choose

For the daily "render me a poster" job, the base 3.0 node is enough and cheaper. Pro earns its premium when the render is the deliverable: the final poster, the hero image, the shot where the text must be letter-perfect and the composition rich. The honest workflow is to iterate on the cheaper tier and promote the winning prompt to Pro for the final pass - the same pattern that works across the whole EvoLink pack, and the one that keeps the meter from eating your balance. As with any cloud call, prompts and reference images leave the machine, and per-call pricing means the cost math is on you. For a closed model with no local door, that's the deal - just don't pay the Pro premium while you're still deciding what to render.