Qwen Image 2.1 reference editing: target order, reference count and output size

File and download boundary: ComfyUI World provides source references and metadata only. Model and LoRA weights, node packages, and workflow files stay with their original providers; this site does not store, mirror, or proxy them. Downloads and access conditions are controlled by each provider.

Start with one target image

The official edit entry is image_qwen_image_2_1_image_edit.json. This review read its header and author link, not every connection. The behavior below is grounded in the core node implementation and the separately reviewed background-removal template. It is not a claim that the complete edit graph has been executed here. Q06 Q04 Q07

Make image order explicit

Place the image to change in image_1, then add supporting subjects or objects as references. State their roles using markers such as <image1> and <image2>. An original test instruction is: In <image1>, replace the blue bag with the bag in <image2>. Keep the person's pose and the background unchanged. This is a proposed test, not a demonstrated output. Q07

Distinguish interface capacity from documented support

Qwen documents up to ten reference images. The ComfyUI 0.37.0 schema provides names for as many as sixteen inputs. That difference does not establish sixteen-image quality or reliability. Keep the tutorial baseline within the author's documented ten-image scope. Q01 Q04

Interpret resolution as a sizing control

The implementation uses the square of the resolution value as an approximate pixel budget, preserves aspect ratio and rounds dimensions to multiples of 32. A value of 1024 therefore corresponds to roughly one megapixel, not 1,024 total pixels. A value of zero retains near-original image dimensions with block alignment, which can make large references expensive. The first valid reference determines the latent canvas returned by the node. Q04

Preserve the edit canvas

Use the encoding node’s latent or follow the template’s explicit custom-size path. The background-removal template warns that replacing the canvas with a poorly matched size can shift an edit. Q07 Intentional reframing is a separate experiment from preserving the original composition.

Verify incremental changes

Begin with a small visible edit on one image. Add a second reference only after recording that baseline. Check identity, untouched regions, reference-role assignment and actual output dimensions. Keep a separate record for each added reference. Instructional masks or annotations should not be sold as pixel-lock guarantees: workflows that require unchanged pixels need an explicit compositing and difference-check stage.

Sources and verification

Source and code review on 2026-09-22. No GPU run, weight download or quality benchmark. Procedures are editorial guidance, not a confirmed diagnosis of your environment.

• Q01 · Qwen-Image-2.1 official README
• Q04 · ComfyUI 0.37.0 Qwen nodes
• Q06 · Official Qwen 2.1 image-edit template
• Q07 · Official Qwen 2.1 background-removal template

Related reading

• Qwen Image 2.1 in ComfyUI: models, workflows and release boundaries
• TextEncodeQwenImage21: reference inputs, resolution and latent output
• Qwen Image 2.1 memory troubleshooting by execution stage
• Qwen Image 2.1 transparent PNG: inspect alpha, not the checkerboard