Prepare Genjutsu character, outfit and product references
Build a clear reference set, distinguish subject and scene roles, and inspect imported inputs before your next remix.
A reference set is a small casting brief. Each image should have a purpose you can explain: the performer’s identity, an outfit, a product, or a location. Similar-looking images with unclear roles can make the intended change harder to communicate.
Choose the image for the job
For a character, use a view that makes the face and silhouette easy to distinguish. For clothing, show the garment rather than hiding it behind a pose. For a product, keep the object large enough to inspect its outline and relevant surface. A scene reference should communicate the intended space and lighting without being mistaken for another subject.
These are preparation suggestions. More references do not automatically produce a more faithful result. Begin with the smallest set that describes your brief, then add a view when a specific ambiguity needs resolving.
Write a reference map
Keep this short map beside your draft:
image 1 -> performer
image 2 -> jacket
image 3 -> bottle
image 4 -> location
An original instruction could be:
Use image 1 for the central performer, image 2 for their jacket and image 3 for the bottle in their right hand. Use image 4 only as the location reference. Follow the source performance.
This suggested instruction has not been generation-tested. Check whether your selected mode matches the intended change using the mode guide.
Review inputs in the workspace
OpenFlux 3 accepts an ordered set of 1–8 reference images for these two official API modes. Inspect each thumbnail after upload. A failed upload or missing image needs correction before submitting. When importing a preset, compare the loaded order with its prompt; inserting or removing a reference can change every later number.
The public API limit is different from the larger reference set visible in some upstream channel examples. A preset marked reference-only remains useful to inspect. A separately labeled adapted version is a revised brief, not a promise to reproduce the original result.
Before you add another image
Identify the missing information. Is the face obscured? Is the product’s reverse side important during a turn? Is the outfit confused with the background? Add a useful view for that question and adjust the reference map. Avoid changing the entire set when one role needs clarification.
Keep a copy of the exact ordered inputs with each attempt. You can prepare them in Motion Transfer, Object Swap, or by reusing a complete preset. Reuse loads a draft and does not generate automatically.
The input-count boundary is documented in the official Motion Transfer API reference. The preparation methods above are OpenFlux 3 editorial guidance.