Nodes/Eric_Qwen_Edit_Experiments/Eric Qwen-Edit Multi-Image Fusion
ComfyUI Node

Eric Qwen-Edit Multi-Image Fusion

A ComfyUI node in Eric Qwen-Edit with 23 inputs and 1 output.

By EricRollei·Created 5 months ago·Updated 3 months ago· 19
Eric Qwen-Edit Multi-Image Fusion
  • pipeline
  • image_1
  • image_2
  • image_3
  • image_4
  • image
promptAll subjects are sitting together on a couch, looking at the camera.
composition_modegroup
subject_labelperson
main_imageimage_1
vae_target_size0
vl_1true
vl_2true
vl_3true
vl_4true
ref_1true
ref_2true
ref_3false
ref_4false
negative_prompt
steps8
true_cfg_scale4.0
seed0
max_mp8.0
CategoryEric Qwen-Edit

Inputs (23)

NameTypeDefaultDescription
pipelineQWEN_EDIT_PIPELINE
image_1IMAGEFirst input image (Picture 1) — default main image
image_2IMAGESecond input image (Picture 2)
promptSTRINGAll subjects are sitting together on a couch, looking at the camera.In 'raw' mode: write the full prompt referencing Picture 1, Picture 2, etc. In other modes: describe the desired scene/action (image references are auto-added).
composition_modeCOMBOgroupgroup: place subjects side-by-side with positions | scene: use Picture 1 as background, place others into it | merge: fuse features into one subject | raw: your prompt exactly as-is
image_3optIMAGEOptional third image (Picture 3)
image_4optIMAGEOptional fourth image (Picture 4)
subject_labeloptSTRINGpersonWhat to call each subject in auto-generated prompts (e.g. 'person', 'woman', 'character', 'bear', 'product')
main_imageoptCOMBOimage_1Which image is the primary reference. Its VAE latent seeds the denoising process for strongest reconstruction.
vae_target_sizeoptINT00–2048Fixed resolution for VAE encoding. 0 (default) = encode refs at output resolution (matches Edit node behavior, best for high-res). Set to e.g. 1024 to force all refs to ~1MP (only useful at low output res).
vl_1optBOOLEANtrueInclude image_1 in VL/semantic path (text encoder understands its content)
vl_2optBOOLEANtrueInclude image_2 in VL/semantic path
vl_3optBOOLEANtrueInclude image_3 in VL/semantic path
vl_4optBOOLEANtrueInclude image_4 in VL/semantic path
ref_1optBOOLEANtrueInclude image_1 in VAE/ref path (pixel-level latent reference)
ref_2optBOOLEANtrueInclude image_2 in VAE/ref path
ref_3optBOOLEANfalseInclude image_3 in VAE/ref path (default False — VL-only for secondary images)
ref_4optBOOLEANfalseInclude image_4 in VAE/ref path (default False — VL-only for secondary images)
negative_promptoptSTRINGWhat to avoid in the output
stepsoptINT81–100Inference steps (8 for lightning LoRA, 50 for base model)
true_cfg_scaleoptFLOAT4.01–20True CFG scale (main quality control)
seedoptINT00–18446744073709550000Random seed
max_mpoptFLOAT8.00.5–16Max output megapixels. VAE refs scale to match output resolution.

Outputs (1)

NameTypeDescription
imageIMAGE