Qwen-Image-Edit-Plus

Qwen-Image-Edit-Plus is the upgraded sibling of the standard Qwen-Image-Edit node - same family, same "describe the change" workflow, but tuned for the jobs where the base editor gets loose. Where the original re-emits the whole frame and lets unrelated pixels drift, the Plus line is the one you reach for when you care about keeping the subject intact while changing the scene around it. If you've ever edited a character into a new background and watched their face subtly mutate, this node is the fix.

Think of the two nodes as tiers, not rivals: plain Qwen-Image-Edit for quick, surgical single instructions, and Plus when you want higher fidelity to the reference - more consistent subject identity, cleaner large-scale changes. It still won't do a chain of ten edits without drift (nothing does), but it gets noticeably more of them right on the first pass.

What you set

• referenceImages (required) - the image(s) to edit, as IMAGE.
• positivePrompt (required) - the instruction in plain language.
• width / height (512–2048 - note the higher floor than the base editor, Plus expects a real canvas).
• steps - off by default; let the model pick. There's no aggressive default here, which is the distilled-editor norm.
• CFGScale / settings.trueCFGScale - the now-familiar pair: CFG off, trueCFG only if the model ignores you.
• lora, ultralytics, watermark, acceleratorOptions sockets, negativePrompt (ignored at CFGScale ≤ 1), seed, numberResults (up to 20), outputFormat, outputQuality - the standard spread.

Output is image (IMAGE).

How it works

Same pack machinery - request with model AIR + taskType over REST, IMAGE downloaded back, cost on the title bar. The difference from the base node is the model behind it: a newer, heavier edit checkpoint that trades a little speed for better adherence to your reference and instruction. Both variants run in the cloud, so "heavier" here just means a slightly longer queue and a slightly bigger bill, not a hardware problem on your end.

Installing

ComfyUI Manager → search Runware → install → restart. Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/Runware/ComfyUI-Runware
pip install -r ComfyUI-Runware/requirements.txt

No model downloads; dependencies are runware-sdk, pillow, soundfile. API key via Settings → Runware API key, RUNWARE_API_KEY, or runware auth login.

Troubleshooting

Set the API key before anything else - it's the #1 failure mode for the whole pack. If Plus still drifts on a hard edit, the tooltip advice holds: keep instructions surgical and don't chain changes. And if a prompt works fine in the base editor but fails here, remember the two models are different checkpoints - Plus has its own sweet spot for phrasing, so rephrase rather than assuming a bug.