Flexible Image Roles for Video Generation

While this request was already included as a side request within the functional request for a ⟨⟨Reference-Guided Video Continuation feature⟩⟩ (posted 2026-09-17), let me raise it independently here for clarity so that it may not be missed.

For Wan 3.0 and MiniMax H3 related models, users currently have to choose between “Reference to Video” and “Image to Video.” The former accepts reference images, whereas the latter allows a start frame and an end frame. While this separation may be inherited from the native models, SeaArt could potentially provide a more flexible UI layer on top of it.

A simple approach would be to let users upload the images for a planned clip first and then assign a role to each image individually. By default, an uploaded image could be treated as a reference frame; hovering over or selecting its thumbnail could expose the three core roles for reassignment where needed: Reference frame | Start frame | End frame.

There may be multiple reference frames, whereas the start frame and end frame, if used, must obviously each be unique. (As an optional extension, a Loop frame role could also be useful for scenes intended to loop.)

This distinction between at least the above three core roles matters when the user has a particular visual progression or plot in mind. Within current Reference to Video workflows, explicitly instructing the model to treat one of the uploaded reference images as the start or end frame typically does not reliably impose that structural role.

Feature request in a nutshell:
Allow flexible role assignment for uploaded images within a single video generation instead of requiring the user to choose between reference-image and start/end-frame modes.

Core roles: Reference frame | Start frame | End frame
Optional extended role: Loop frame