
Two film-strip concepts suggesting a source clip restyled into a new scene
Video-to-video is for when you already have motion you trust. You keep the performance or camera idea and ask for a new look, setting, or continuation. FLUX 3 lists video-to-video, continuation from existing video and audio, and keyframe-controlled transitions as part of the early video toolkit.
Note: On TheFluxTrain you can extend a clip and interpolate start/end or keyframes. BFL also announced a broader restyle video-to-video mode.
Quick answer: FLUX 3 can continue from existing video, or fill motion between start and end frames. Native audio stays in the picture. On TheFluxTrain that is extend plus interpolation. Broad editing APIs and private weights are still a BFL plan.
Pillar guide: FLUX 3 image, video, editing, and release status.
From BFL's announcement, treat these as related video control modes:
This is generative control over clips. It is not a full NLE timeline editor.
Use it when:
Prefer image-to-video when you only have stills. Prefer text-to-video when you are exploring from scratch.
Keyframes give you control over transitions. You mark important visual moments and ask the model to fill the motion between them. That helps for title cards, map moves, or staged product reveals where the endpoints matter more than the middle.
Until you try the UI, assume keyframe workflows need clear endpoint images and a duration inside the 20-second limit. On TheFluxTrain, start and end stills live on Image Interpolation.
Parts of the design fit editing tasks: restyle, continue, reference carry-over, planned API editing access. A generally available, fully specified editing product is not here yet. The careful answer lives in the pillar post under Is FLUX 3 an AI image and video editor?.
Keep the woman walking left to right and the camera pan. Change the street to a rainy night market in Taipei. Keep footsteps. Replace dialogue with Mandarin asking for directions. 6 seconds.
Yes, as an announced FLUX 3 Video capability.
BFL lists continuation from existing video and audio.
Video and audio generation and editing are planned through APIs and private-weight access. Broad access is not available yet.
Image-to-video starts from stills. Video-to-video starts from motion you already have.
FLUX 3 Video is built around native audio. Exact behavior when continuing or replacing sound will need hands-on confirmation.
Extend and interpolation are live. A separate restyle-only page is not.
FLUX 3 Dev is announced as a future open-weight release with no date. License and hardware needs are unpublished.
Start with the release overview, then this page for clip-based workflows.