SF WaveSpeed Qwen Edit

Image editing is where local models still struggle and hosted ones shine, and Qwen-Image-Edit is one of the strongest hosted editors around: a 20B MMDiT built on Qwen-Image that edits while preserving the original style. SF WaveSpeed Qwen Edit is the wrapper that puts it in your ComfyUI graph - give it an image URL and an instruction, get back an edited image.

The genuinely useful trick here is the bilingual thing. Qwen-Image-Edit was trained for Chinese and English instructions, so it handles both naturally - handy if you're working with prompts in either language, and it's better than average at understanding edits that involve written text in the image (signs, labels, UI mockups). If you've ever fought with an editor that garbles in-image text, this one is worth a try.

It's part of the comfyui-stillfront pack, an API call through WaveSpeed - no local weights.

How it works

You hand the node a URL to the source image and an edit prompt. It posts both to WaveSpeed's qwen-image/edit endpoint, downloads the result, and returns an IMAGE tensor. The heavy lifting - understanding what "change the shirt but keep the lighting" means and rendering it - happens on their servers.

The inputs that matter:

• prompt - your edit instruction, Chinese or English. The tooltip's example style is what works: plain, specific language about what changes.
• image_url - the source image as a URL string, connected from the Upload Image node (the one that gives you a hosted URL). This is not an IMAGE tensor input.
• seed - -1 for random, or lock it for reproducible edits.
• output_format - jpeg/png/webp.
• enable_sync_mode - default on; the node waits for the result and returns the image.

Output is a single image (IMAGE).

Installing it

Pack install plus a WaveSpeed key:

cd ComfyUI/custom_nodes
git clone https://github.com/razvanmatei-sf/comfyui-stillfront
cd comfyui-stillfront
pip install -r requirements.txt

Restart, then set your key in the SF WaveSpeed Client node wired into this node, in config.ini, or via WAVESPEED_API_KEY.

Troubleshooting

• "The image_url input won't connect." It takes a URL string, not an IMAGE wire. Connect the URL output of an Upload Image node - the tooltip spells this out, and it's the most common first-run stumble with all the edit nodes in this pack.
• Edit comes back but changed too much. Qwen-Image-Edit is a full generator at heart; strong instructions (or a high-quality source that contradicts the prompt) make it drift. Restate the edit more conservatively - "keep everything the same except…" phrasing helps.
• In-image text still garbled. It's better than most at text, but it's not OCR-perfect. Keep text edits short.
• Auth errors. Same key story as every WaveSpeed node: client node, then config.ini, then env var.

Worth noting: as a hosted model, every edit is billable and subject to WaveSpeed's API content policy (which the community finds more permissive than the web UI). And if you want to steer the edit with a LoRA, the LoRA variant of this node is the natural next step.