blackforestlabs / flux-3-i2v

## Overview FLUX 3 I2V generates short videos from still images plus a written scene direction. Black Forest Labs trained FLUX 3 as one multimodal model, so it can plan motion and generate synchronized audio in the same pass. You can animate a single image, interpolate between a first and last frame, or lock a shot to multiple keyframes. This is useful when you need controlled motion from specific visuals, without manual animation or video editing. ## What you can build - Animate a product photo into a short ad shot with camera motion and sound design cues - Turn concept art into motion tests for storyboards and pitch decks - Create first-to-last transformations like outfit changes, aging, or scene transitions - Make keyframed explainer clips where each slide-like moment must land on time - Generate social clips from a hero image in vertical, square, or cinematic ratios ## Inputs - A required written prompt that describes the scene, action, camera behavior, and (if wanted) the sounds you expect to hear - Optional reference images (PNG, JPEG, or WebP), using 1 to 10 images total - 1 image: it becomes the starting frame that gets animated - 2 images: they become the first and last frames, and the model generates the transition - 3–10 images: they become keyframes that the video must hit during the shot - Optional keyframe timing for the images between the first and last image, written as comma-separated seconds in increasing order - These timestamps map onto a 24 fps timeline, so each keyframe must fit inside the full duration - A required clip duration in whole seconds, from 5 to 20 - An optional aspect ratio choice, including auto, 21:9, 2:1, 16:9, 4:3, 1:1, 3:4, and 9:16 - A required output resolution choice: 720p or 1080p - An optional toggle to generate audio or return a silent video - An optional safety strictness level from 0 (strictest) to 4 (most permissive) ## Outputs - An MP4 video file you can download and use in editing tools - Video frames rendered at 24 fps - If audio is enabled, the MP4 includes an audio track generated with the visuals - A numeric seed value that identifies the specific generation result ## Limitations - Clip length is limited to 5–20 seconds. - Output resolution is limited to 720p or 1080p. - The frame rate is fixed at 24 fps. - Keyframed workflows support up to 10 images, and each keyframe must map to a valid moment inside the duration. - If you don’t name the sounds you want, quiet scenes can come back with little audible detail. - Low-quality inputs can reduce control. Blurry images, heavy compression, tiny subjects, or inconsistent lighting can cause drift between frames. - Mixed styles across keyframes can produce unstable motion. Keep keyframes visually consistent for best continuity. ## Safety & compliance - The model applies safety filtering. Higher strictness can block more requests and outputs. - Don’t upload images you don’t have rights to use, especially faces, brands, or copyrighted characters. - Avoid requests that involve illegal activity, sexual content featuring minors, or real-person harm. - Follow your platform’s content rules when generating, sharing, or monetizing outputs.