Black Forest Labs: FLUX 3 Video

FLUX.3 Video is a video generation model from Black Forest Labs, the same provider behind the FLUX image models. It sits in the video category and is built for producing video rather than text, so it fits pipelines where the final artifact is a clip. If you already use Black Forest Labs models elsewhere in your stack, this extends that same line of work into motion.

On the input side, the model accepts text, image, and video, and it returns video. That combination covers text-to-video, image-guided generation using an opening frame, a closing frame, or both, and video continuation workflows where you feed existing footage in and extend it. Output can be generated at 720p or 1080p, with aspect ratios of 21:9, 16:9, 4:3, 1:1, 3:4, and 9:16, so both wide cinematic framing and vertical social formats are available. Clip durations run from 5 to 20 seconds in one-second steps, and audio generation is supported. No context length is published for this model, which is expected for a video generator.

The practical uses follow from those controls. Keyframe support lets you pin the start and end of a shot, which is useful for storyboard-to-clip work, product shots that must open and close on specific frames, and transitions where you need a defined landing point. Video continuation is the natural fit for extending existing footage, turning a short take into a longer sequence, or building out a scene from a clip you already have. Aspect ratio and resolution options make the same model usable for widescreen edits at 1080p and for vertical output at 720p without switching tools. Audio generation means you can produce a clip with sound in one pass rather than muxing it afterwards.

You call it through api14's OpenAI-compatible API using the model id "black-forest-labs/flux-3-video". No request parameters are currently listed as supported for this model, so the input types (text, image, video) and the output settings described above are what define a request rather than a tunable parameter set. Because the endpoint follows the OpenAI-compatible format, existing client libraries and request shapes you already use for other models on api14 apply here.

The listed maximum output is 0, and no context length is published, so treat generation settings as the things that define a request rather than token budgets. Durations cap at 20 seconds, which means longer sequences need continuation or stitching across multiple calls. If your task is text generation, image generation, or anything outside video, a model in the matching category on api14 is the better choice.