Black Forest Labs announced FLUX 3 on July 23, 2026. It is the biggest change to the FLUX model family so far: instead of focusing only on image generation and editing, FLUX 3 is a multimodal foundation model trained across images, video, audio and action.
FLUX 3 is now available on PoYo through five production video workflows: text-to-video, image-to-video, first-and-last-frame generation, video extension, and positioned keyframes. The FLUX 3 API page on PoYo includes the live generator, model IDs, supported controls, and pricing.
This guide separates what Black Forest Labs has officially announced from the details that remain unknown.
FLUX 3 at a Glance
| Feature | Officially announced information |
|---|---|
| Developer | Black Forest Labs |
| Announcement date | July 23, 2026 |
| Current status | Five production video workflows available on PoYo |
| Modalities | Image, video, audio and action prediction |
| Inputs | Text, images and video references |
| Image capabilities | Generation and editing |
| Video capabilities | Text-to-video, image-to-video, video-to-video, continuation and keyframes |
| Native audio | Yes |
| Maximum announced video duration | Up to 20 seconds in one generation |
| Multilingual dialogue | Announced |
| General PoYo API access | Available now |
| Public API model IDs and pricing | Published in the PoYo documentation and model page |
What Is FLUX 3?
FLUX 3 is a unified multimodal model. Black Forest Labs says it was trained on images, videos and audio at the same time so that the modalities constrain and inform one another.
An image describes spatial structure. Video adds time and motion. Audio provides evidence about events and physical interactions. Language connects those signals to instructions and abstract goals. FLUX 3 is designed to learn a shared representation rather than treating each medium as an isolated generation task.
This is a different product direction from FLUX 2, which is primarily an image generation and editing family. FLUX 3 still generates and edits images, but it also produces video with native audio and provides a foundation for action prediction.
The FLUX 3 Model Family
Black Forest Labs has announced four major applications of the FLUX 3 foundation.
FLUX 3 Image
FLUX 3 Image covers image synthesis and editing through APIs and private weight access. BFL says the model handles varied styles, aspect ratios and resolutions, follows complex prompts more accurately and improves multilingual text rendering.
FLUX 3 Video
FLUX 3 Video generates and edits video with native audio. It accepts text, images and video references and supports several workflows, including continuation and keyframe-controlled generation.
FLUX 3 Dev
FLUX 3 Dev is the announced open-weight multimodal backbone. BFL describes it as a foundation for content creation and action prediction. The company has not yet published its weights, license, hardware requirements or release date.
FLUX 3 Action and FLUX-mimic
FLUX 3 also supports action prediction through selected research and commercial partners. FLUX-mimic, developed with mimic robotics, applies the model's visual world representation to robot actions. This is a specialized research and industrial direction, not a consumer image or video generator.
FLUX 3 Image Capabilities
FLUX 3 remains a major image model even though its scope is broader than earlier FLUX releases.
Black Forest Labs has announced:
- Text-to-image generation
- Image editing
- Multiple visual styles and aspect ratios
- Multiple output resolutions
- Better handling of complex prompts
- More accurate multilingual text generation
- API and private-weight access
The official examples range from photorealistic close-ups and product-like studio images to paintings and flat illustrations. They demonstrate style diversity, but they are selected launch samples rather than an independent measure of consistency.
Important production details are still missing. BFL has not published the final FLUX 3 Image API parameters, maximum resolution, latency, rate limits or price.
FLUX 3 Video Capabilities
FLUX 3 can generate videos with audio up to 20 seconds long in a single generation. The announced workflows include:
- Text-to-video: create a complete clip from a written description.
- Image-to-video: animate a starting frame or use images as visual references.
- Video-to-video: carry characters, subjects or visual elements from a source clip into another scene.
- Video and audio continuation: extend an existing audiovisual sequence.
- Keyframe-to-video: generate controlled transitions between defined moments.
- Multilingual dialogue: generate spoken content in multiple languages.
- Multi-shot chaining: connect clips into longer sequences using an agentic workflow.
- Animated typography: produce motion design and animated text.
The model's audio is native rather than added as a separate post-processing step. This matters for events where sound, motion and timing need to agree.
How Self-Flow Fits into FLUX 3
FLUX 3 builds on Self-Flow, a method developed by Black Forest Labs for aligning multimodal generation and understanding in one architecture.
BFL's preliminary research compares Self-Flow with conventional flow matching across image, video and audio generation. The company reports lower normalized generation error and faster learning on downstream manipulation tasks.
These results help explain the architecture, but they should not be interpreted as a complete public benchmark of the final FLUX 3 products. The model and evaluation harness are still being developed.
Early FLUX 3 Video Evaluations
Black Forest Labs published preliminary human-preference results for 10-second, 720p text-to-video clips with audio. In its tests, FLUX 3 was preferred over:
- Grok Imagine Video in up to 69% of comparisons
- Kling v3 Pro in 60%
- Happy Horse v1 in 59%
- Happy Horse 1.1 in 57%
- Seedance 2.0 and Gemini Omni Flash in 52%
- Runway Gen-4.5 in 77%
- Luma Ray 3.2 in 93%
These are vendor-published early evaluations, not an independent leaderboard. BFL explicitly says the model and evaluation setup are still evolving. Prompt selection, sampling settings, evaluators and competing model versions can all affect preference results.
FLUX 3 Availability on PoYo
PoYo currently exposes five production video model IDs:
flux-3/text-to-videoflux-3/image-to-videoflux-3/first-last-frame-to-videoflux-3/extend-videoflux-3/keyframes-to-video
These workflows generate 5–20 second clips at 720p or 1080p, support eight aspect ratios, and can create synchronized native audio. Standard workflows cost 34 credits per second at 720p or 58 credits per second at 1080p. Video extension costs 82 or 106 credits per second respectively.
This launch applies to PoYo's FLUX 3 video workflows. FLUX 3 Image, Action, and the open-weight FLUX 3 Dev backbone remain separate parts of the broader BFL roadmap; PoYo does not present them as included in the current video API release. See the FLUX 3 API documentation for current request fields and limits.
FLUX 3 Use Cases
Image generation and editing
FLUX 3 Image may support product visuals, editorial illustration, advertising, multilingual design and reference-controlled editing.
Short videos with synchronized sound
Native audiovisual generation is useful for dialogue scenes, product demonstrations, social clips, animated designs and story concepts where sound must match the action.
Character and style continuity
Image and video references can carry a central subject or visual language into new scenes. Multi-shot chaining could extend this into longer sequences.
Pre-visualization
Keyframes, reference clips and image inputs can help creative teams explore transitions, camera movement and scene structure before conventional production.
Physical AI research
FLUX 3's shared world representation can also be adapted for action prediction, as demonstrated by the FLUX-mimic collaboration.
Frequently Asked Questions
Is FLUX 3 released?
Black Forest Labs announced FLUX 3 on July 23, 2026. PoYo now provides five production video endpoints covering text, images, first and last frames, source-video extension, and positioned keyframes.
Is FLUX 3 an image model or a video model?
It is a multimodal foundation model. Announced variants cover image generation and editing, video with native audio, and action prediction.
Does FLUX 3 generate audio?
Yes. FLUX 3 Video generates native audio together with video and supports multilingual dialogue.
How long can FLUX 3 videos be?
PoYo accepts integer durations from 5 through 20 seconds and supports 720p and 1080p output.
Is FLUX 3 open source?
BFL has announced an open-weight FLUX 3 Dev backbone, but the weights, license and release date have not yet been published.
Is the FLUX 3 API available on PoYo?
Yes. Use the FLUX 3 model page or integrate one of the five documented PoYo video model IDs.
How is FLUX 3 different from FLUX 2?
FLUX 2 focuses on image generation and editing. FLUX 3 expands the family into a unified model for images, video, native audio and action prediction. See the detailed FLUX 3 vs FLUX 2 comparison.
Final Takeaway
FLUX 3 is not simply another image-model upgrade. It changes FLUX from an image-focused family into a multimodal platform designed to connect visual appearance, motion, sound and physical behavior.
The most interesting capabilities are clear: native audiovisual generation, reference-driven video, multilingual dialogue, stronger typography and a shared architecture that extends toward action prediction. PoYo now makes the five released video workflows available with documented model IDs, controls, and per-second pricing.
Use the PoYo FLUX 3 model page to generate video now, or read the API documentation before integrating. Continue to treat unreleased Image, Action, and Dev details separately from the video endpoints available today.
Sources: Black Forest Labs FLUX 3 announcement, FLUX 3 x mimic