Redux迁移
FLUX-Redux-style object migration, fully hosted
- migrate_image
- migrate_mask
- Product_image
- Product_mask
- output
Display name "Redux迁移" - "Redux migration" - and the family resemblance is intentional: this is the pack's hosted take on FLUX Redux-style object migration. One picture has an object you like (a specific handbag, a watch, a pair of shoes), another picture is where you want it to end up (a model on a sofa, a product scene). You feed both, plus a mask marking the object, and the backend moves it over. The prompt the node sends says it all: "This is a collage picture - the left objects replace the objects in the right picture."
What makes this one worth a look is how it squares with the pack's whole pitch: everything configured, no model downloads. The README's English section is explicit that this is a FLUX Redux migration made ready-to-run. You're not cloning Flux, not downloading Redux adapters, not babysitting a diffusion_models folder - the migration happens server-side on Mojie's infra and the finished composite comes back as an IMAGE.
The inputs that matter
migrate_image+migrate_mask- the object you're migrating, plus the mask that marks its region. Both required. If you've ever done Redux locally you know the drill: the mask tells the model what to lift out.Product_image- the target scene. Required.Product_mask- optional, and the author's own docs flag it: it's not mandatory, but adding a mask for the target region makes detail restoration noticeably more accurate. When the replacement has to respect the original scene's lighting and perspective, that extra mask earns its keep.prompt- optional; there's a default built in. Use it to say what the object is ("a brown leather handbag") when the model needs a hint.strong- the migration strength, default 0.6. The author's guidance: 0.6–0.9 is the working range. Below that the swap half-heartedly merges; above it you start to lose the target scene.seed- for variety across runs.
Output is output, a single IMAGE.
Installing it
One repo, one key, same as its packmates. Search "mojieapi_party" in ComfyUI Manager, or:
cd ComfyUI/custom_nodes
git clone https://github.com/MoJIeAIGC/comfyui-MJAPI-party.git
Register at mojieaigc.com, paste the key into config.ini (leave BASE_URL alone), restart ComfyUI. No model files. Dependencies are minimal - the chunky one is opencv-python - and there's a settings panel in ComfyUI for updating the pack later.
Common issues
- Red error image on output. That's the pack's error channel - a 401 means the key, a 403 usually means the balance. Migration calls are metered; they're not the priciest thing in the pack, but they're not free either.
- Weak or mushy swap. Before blaming the model, check
strong. If it's at 0.6 with a complex scene, that's a "subtle blend" more than a "hard replace." And if the target area wasn't masked, the model is guessing where the object should land - addProduct_mask. - Missing masks. This is the one node in the pack where the mask isn't optional garnish.
migrate_maskis required - forget it and the whole "which object?" question is unanswered. Masking-detection-detailing basics from the KB apply: a clean, tight mask around the object beats a sloppy one every time.
Inputs (7)
| Name | Type | Default | Description |
|---|---|---|---|
| migrate_image | IMAGE | — | |
| migrate_mask | MASK | — | |
| Product_image | IMAGE | — | |
| prompt | STRING | — | |
| strong | FLOAT | 0.60 | — |
| seed | INT | 0 | — |
| Product_maskopt | MASK | — |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| output | IMAGE | — |