FLUX 3 Video is Black Forest Labs' first FLUX model for full audiovisual generation. It creates video and native audio together, accepts text, images and video references, and produces clips up to 20 seconds long in a single generation.
FLUX 3 Video is now available on PoYo through five production workflows with published model IDs, request controls, and per-second pricing. The FLUX 3 API page on PoYo provides a live generator for text, images, first and last frames, source-video extension, and positioned keyframes.
FLUX 3 Video at a Glance
| Feature | Announced support |
|---|---|
| Text-to-video | Yes |
| Image-to-video | Yes |
| Image references | Yes |
| Video-to-video | Yes |
| Video and audio continuation | Yes |
| Keyframe-to-video | Yes |
| Native audio | Yes |
| Multilingual dialogue | Yes |
| Maximum single-generation duration | Up to 20 seconds |
| Aspect ratios | Multiple formats announced |
| Animated typography | Announced |
| Multi-shot sequences | Agentic chaining announced |
| General PoYo API | Available now |
What Is FLUX 3 Video?
FLUX 3 Video is one application of the broader FLUX 3 multimodal foundation model. Unlike an image model adapted to generate frames, FLUX 3 was trained across images, video and audio in a shared architecture.
The goal is to model more than visual appearance. Video teaches motion and temporal relationships. Audio provides information about impacts, materials, speech and environmental events. Joint training allows the generated sound to be connected to what happens on screen.
This is also why FLUX 3 extends beyond content creation into action prediction through specialized research variants.
Official FLUX 3 launch media from Black Forest Labs.
Text-to-Video
Text-to-video creates a clip directly from a written scene description. A useful prompt may define:
- Subject and action
- Location and time
- Camera position and movement
- Visual style
- Spoken dialogue
- Environmental sound
- Timing and sequence of events
- Aspect ratio
FLUX 3 can generate the visual and audio portions in one pass. PoYo requests expose the prompt, duration, resolution, aspect ratio, audio toggle, and the image, source-video, or keyframe inputs required by the selected workflow.
Image-to-Video
FLUX 3 supports two image-driven approaches.
Starting-frame animation
An image can define the first frame, after which the model creates motion and sound. This is useful for animating product images, illustrations, portraits or designed scenes.
Visual reference input
Images can also act as identity, subject or style references without becoming the literal first frame. This can help preserve a character, object or visual direction in a new scene.
On PoYo, image-to-video accepts one source image, first-and-last-frame generation accepts two ordered images, and keyframe generation accepts up to ten positioned images. Check the current API documentation for upload and file-format requirements.
Video-to-Video
Video-to-video uses a source clip as a reference and carries central elements into a different scene or context.
Potential applications include:
- Moving the same character into a new environment
- Reinterpreting footage in another visual style
- Preserving motion while changing the scene
- Creating variations of an existing shot
- Using a performance as a temporal reference
Video-to-video does not guarantee perfect identity or motion preservation. PoYo currently exposes source-video continuation through flux-3/extend-video; it should not be confused with a separate general-purpose video-restyling endpoint.
Video and Audio Continuation
FLUX 3 can continue an input video and its audio. A continuation model must extend several things together:
- Subject appearance
- Camera motion
- Object trajectories
- Lighting and environment
- Dialogue or ambience
- Rhythm and event timing
This workflow could extend short shots or produce consecutive story beats without starting each clip from an unrelated prompt.
Keyframe-to-Video
Keyframe-to-video generates a transition between defined visual moments. Start and end frames can communicate composition more precisely than text alone.
Possible uses include:
- Product transformations
- Camera transitions
- Character movement between poses
- Before-and-after sequences
- Logo or typography animation
- Storyboard-to-motion pre-visualization
PoYo supports up to ten positioned keyframes. Each keyframe supplies an image URL and a frame index, allowing the transition timing to be controlled directly.
Native Audio Generation
Native audio is FLUX 3 Video's defining feature. The model generates sound as part of the same scene rather than relying entirely on a separate audio pipeline.
Event-aligned sound
Actions such as impacts, movement and mechanical interactions can produce sound timed to the visible event. BFL says the model is particularly strong at associating sounds with physical events.
Environmental audio
A scene can include ambience appropriate to its location, weather and activity. That may provide a more complete first draft than silent video generation.
Multilingual dialogue
Black Forest Labs lists multilingual dialogue as a core capability. The launch announcement does not provide a complete language list or measured speech accuracy for each language.
Video and audio continuation
The model can extend an audiovisual input, potentially maintaining ambience, speech context and event timing into the new section.
Native generation does not eliminate post-production. Dialogue accuracy, pronunciation, lip synchronization, audio levels and unwanted artifacts still require review.
Duration and Resolution
PoYo accepts integer durations from 5 through 20 seconds and supports 720p and 1080p output across eight aspect ratios. Native audio can be enabled in the same request.
Standard text, image, first-and-last-frame, and keyframe workflows cost 34 credits per second at 720p or 58 credits per second at 1080p. Video extension costs 82 credits per second at 720p or 106 credits per second at 1080p. Frame rate, codec, generation time, and other operational details should be checked in the current FLUX 3 API documentation.
Character and Style Consistency
FLUX 3 uses visual references to help maintain central elements across scenes. BFL says references can keep characters consistent in multi-scene sequences lasting several minutes.
That longer output is created by chaining individual clips, not by generating several minutes in one call. Each shot may still require:
- Reference selection
- Prompt continuity
- Transition planning
- Identity review
- Audio continuity checks
- Editing and assembly
Agentic chaining may automate part of the planning, but it does not turn the 20-second single-generation limit into an unlimited continuous shot.
Typography and Animated Design
BFL highlights strong typography and animated design as FLUX 3 Video capabilities. This could support:
- Motion posters
- Title sequences
- Product announcements
- Kinetic typography
- Social advertisements
- Interface and graphic animation
Generated text should still be checked frame by frame. A title may look correct in one moment and deform during motion.
Early Evaluation Results
In BFL's preliminary human-preference tests, FLUX 3 was compared with several video models. The company reports preference rates ranging from 52% against Seedance 2.0 and Gemini Omni Flash to 93% against Luma Ray 3.2.
These results require context:
- They were published by the model developer.
- They used 10-second, 720p text-to-video clips with audio.
- The FLUX 3 candidate and evaluation harness were still in development.
- “Up to” results may summarize subsets rather than every prompt category.
- Independent reproduction is not yet available.
The numbers are useful early evidence, not proof that FLUX 3 is universally better.
Practical Use Cases
Dialogue-driven short scenes
Native speech and facial expression make FLUX 3 relevant to short narrative moments and multilingual character concepts.
Product and brand videos
Reference images can establish a product or visual identity, while native sound creates a more complete campaign draft.
Social video
Multiple aspect ratios, animated text and short duration fit advertisements, announcements and vertical social formats.
Pre-visualization
Images, reference video and keyframes can turn storyboards into moving previews for camera, timing and transition decisions.
Video variation and continuation
Teams can explore alternate scenes, extend a shot or carry a central character into another context.
Multi-shot storytelling
Chained clips may create longer sequences while visual references help maintain consistency.
FLUX 3 Video API Availability
FLUX 3 Video is available now on PoYo through five production model IDs:
flux-3/text-to-videoflux-3/image-to-videoflux-3/first-last-frame-to-videoflux-3/extend-videoflux-3/keyframes-to-video
All five use PoYo's asynchronous generation flow for task submission and status retrieval. Use the FLUX 3 model page for interactive generation or the API documentation for integration details.
Frequently Asked Questions
Is FLUX 3 Video available now?
Yes. PoYo provides five production workflows for text, images, first and last frames, source-video extension, and positioned keyframes.
Does FLUX 3 Video generate sound?
Yes. It generates native audio together with video, including environmental sound and multilingual dialogue.
How long can a FLUX 3 video be?
BFL says a single generation can be up to 20 seconds. Longer sequences can be assembled by chaining clips.
Does FLUX 3 support image-to-video?
Yes. It can animate a starting frame or use images as visual references.
Does it support video-to-video?
Yes. BFL says a source video can guide central elements such as a character in a new scene or context.
Can FLUX 3 continue an existing video?
Yes. Video-and-audio continuation is one of the announced workflows.
Which languages does FLUX 3 support?
BFL announced multilingual dialogue but has not published a complete supported-language list.
What does the FLUX 3 Video API cost?
Standard workflows cost 34 credits per second at 720p or 58 at 1080p. Video extension costs 82 credits per second at 720p or 106 at 1080p.
Conclusion
FLUX 3 Video combines several workflows that previously required separate models: text-to-video, image animation, video transformation, continuation, keyframe transitions and audio generation.
Its most important difference is not only the 20-second duration. It is the attempt to generate motion and sound from one shared representation, while accepting references across multiple media.
Read everything currently known about FLUX 3, compare FLUX 3 and FLUX 2, or use FLUX 3 on PoYo.