FLUX 3 Video is now available in ComfyUI through Partner Nodes. Video generation is what ships today. Open weights are coming soon.
Black Forest Labs’ first unified multimodal model — one model for video, audio, image, and action-prediction, built on a richer understanding of how the world looks, moves, sounds, and how to interact with it.
The training is the reason. Rather than separate systems stitched together at inference time, all four modalities are trained jointly in one architecture, and each modality sharpens the others. What comes out is a true general model: creations that are truer to life, stylistically diverse well beyond cinematic, and deeply customizable.
Model Highlights
FLUX 3 Video generates up to 20 seconds in a single generation, with audio produced natively alongside the frames rather than dubbed on afterward — dialogue, sound effects, and ambience in the same pass, and optional when you want silent plates to score yourself. Stylistic range is the differentiator. Most video models converge on a cinematic aesthetic regardless of what you ask for; FLUX 3 makes any style possible, from hyper-realistic to animation to cinematic.
What that covers in practice:
Text to video — Simple prompts or complex ones, with real world knowledge behind the details.
Image to video — Drive from a start frame, an end frame, or a sequence of keyframes in order.
Video continuation — Extend an existing clip while carrying momentum, framing, and scene logic forward.
Multiple scenes — Several scenes and camera angles inside a single generation.
Multilingual dialogue — Accurate speech across languages and accents, with lipsync that holds.
Text and typography — Type that sits naturally in the scene rather than pasted on top.
Agentic chaining — Generate and Chain clips together in ComfyUI for a longer, coherent story.
The same architecture reaches past video. Image generation, fast variants, and open weights are on the way.
Getting Started
Update ComfyUI to the latest version, or open Comfy Cloud.
Download the workflows below, or find them in the template library.
Write your prompt, upload an image, set duration, and run.
FLUX 3 joins the growing roster of models available through ComfyUI Partner Nodes.
As always, enjoy creating!
The hope that you would continue to develop open‑source models has been dashed...
The start-frame, end-frame and ordered-keyframe options make continuity a concrete test. I would try one reference sequence with a restrained camera move, then compare how the final pose and scene geometry survive a more complex version. I represent the Seedance 2.5 website (https://seedance2-5.co/); reviewing those endpoints is useful for planning an image-to-video shot too, while continuation and keyframe controls should be checked in the particular tool being used.