Black Forest Labs · Video Early Access
FLUX 3 AI Video Generator
Black Forest Labs introduced FLUX 3 on July 23, 2026 as a multimodal foundation model, and video is the FLUX capability that opened first: clips of up to 20 seconds at 720p with native audio. This guide covers the official rollout while the embedded generator runs Kling 3.0, which you can use right now.
FLUX 3 video is in official Early Access, but it is not integrated into SoraVideo. The generator below runs Kling 3.0 today; its outputs are never presented as results from the new model.
Be specific about motion, characters, and scene for the best Kling 3.0 results
Kling 3.0 Demo
Configure your settings and click Generate to create an AI video with Kling 3.0
Recent
Max 20
Sign in to save history
Your generations will sync across all devices
Official model overview
What Is FLUX 3?
FLUX 3 is Black Forest Labs' new multimodal foundation model. Rather than a single-purpose release, it jointly learns from images, videos, and audio inside one architecture, with capabilities spanning video and audio generation, image synthesis and editing, action prediction, and a planned open-weight backbone.
Video is what shipped first. Early Access covers text-to-video, image-to-video, video-to-video, and keyframe-driven generation, producing clips of up to 20 seconds at 720p with native audio and multilingual dialogue. Black Forest Labs also positions the same backbone for action prediction in physical AI.
SoraVideo has not integrated the model. The working form on this page uses Kling 3.0 so visitors can create a clip now without confusing a preview guide with a live integration.
Rollout order
Video Now, Image Next
The staged rollout matters if you are deciding what to plan around. Video Early Access is open today; Black Forest Labs says an Early Access phase for FLUX 3 Image will follow in the following weeks, which is why no production image endpoint exists yet.
For the image side, the announcement highlights synthesis and editing across styles, aspect ratios, and resolutions, with preliminary evaluations reporting stronger complex-prompt handling and multilingual text rendering. Those claims remain mid-training results until image access is public.
This page is updated as each phase opens. If you need finished pictures today, SoraVideo runs released image models such as Z-Image, GPT Image 2, and Ideogram 4.0 on their own tool pages.
Model evolution
From FLUX.1 and FLUX.2 to FLUX 3
FLUX.1 established the FLUX family for high-quality text-to-image generation. FLUX.2 expanded the FLUX product line around professional image generation and editing, including faster and higher-quality variants.
The third generation changes the framing from a still-image generator to a shared model of real-world signals. Motion, sound, and language constrain one another during training, which is what makes video with native audio a first-class output rather than a bolt-on.
For creators, the practical questions stay familiar: prompt adherence, motion stability, audio quality, editability, speed, price, and API reliability. We will update this guide as production documentation matures.
Video Capabilities in Early Access
Official FLUX Early Access details from Black Forest Labs, separated from the Kling 3.0 generator below.
Up to 20-Second Clips
Native Audio & Dialogue
Four Generation Modes
Unified Multimodal Backbone
How to Generate AI Video Today
Use SoraVideo's current Kling 3.0 workflow while official FLUX access rolls out separately.
- 1
Describe the Shot
Write a concrete prompt covering subject, action, setting, lighting, camera movement, and the mood you want the clip to carry.
- 2
Choose Duration and Format
Select an available aspect ratio and generation settings, then review the displayed credit requirement before submitting.
- 3
Generate and Download
Create the clip with Kling 3.0, review the result, and download the finished file. This output is not labeled as coming from the new model.
FLUX 3 vs Kling 3.0, Seedance 2.5, and MiniMax H3
A scope and availability comparison of current AI video models, instead of unsupported winner claims.
| Capability | FLUX 3 | Kling 3.0 | Seedance 2.5 | MiniMax H3 |
|---|---|---|---|---|
| Public status | Video Early Access | Released | Official page, rolling out | Released |
| Max clip length | 20s announced | Documented by Kuaishou | 30s announced | 15s announced |
| Resolution | 720p in Early Access | Documented by Kuaishou | Native 4K announced | 2K announced |
| Native audio | Yes, with dialogue | Yes | Multimodal references | Yes, stereo |
| Also generates images | Announced, opening later | No | No | Yes, multimodal |
| Generator on this page | Not integrated | Runs below | Use its own page | Use its own page |
FLUX facts come from Black Forest Labs' July 23, 2026 Early Access announcement. Other models change frequently; this table compares scope and availability, not output quality rankings.
FLUX Release, Early Access & Availability
FLUX 3 is official and rolling out in stages, starting with video. That does not mean every capability is generally available through a public production API.
- Black Forest Labs announced FLUX 3 on July 23, 2026
- Video Early Access is open — clips up to 20 seconds at 720p with native audio
- Text-to-video, image-to-video, video-to-video, and keyframe modes are covered
- Image Early Access is expected in the following weeks
- The family also covers audio, action prediction, and a planned open-weight backbone
- SoraVideo has not integrated a FLUX 3 route; the embedded form uses Kling 3.0
API & access
No new FLUX 3 backend was added in this release. SoraVideo will only expose a selector after an official access path, pricing model, request schema, provider status flow, credit accounting, and delivered-output checks are complete.
FLUX 3 FAQ
Answers about official status, access, and what the generator on this page actually runs.
Can FLUX 3 generate video?+
Yes — video is the capability that opened first in Early Access: clips up to 20 seconds at 720p with native audio and multilingual dialogue, across text-to-video, image-to-video, video-to-video, and keyframe modes.
Is FLUX 3 officially released?+
Yes. Black Forest Labs announced the model on July 23, 2026 and opened Early Access for video. Early Access is a staged rollout, not general public API availability for every capability.
Is the generator on this page powered by FLUX 3?+
No. The embedded generator currently runs Kling 3.0. SoraVideo has not integrated the new model, and generated clips are not labeled as FLUX 3 outputs.
When can I generate images with FLUX 3?+
Black Forest Labs says an Early Access phase for the image variant will follow in the following weeks. Until then, SoraVideo runs released image models such as Z-Image, GPT Image 2, and Ideogram 4.0.
Is FLUX 3 only a video model?+
No. It is a multimodal foundation model spanning video, audio, image synthesis and editing, action prediction, and a planned open-weight backbone. Video simply shipped first.
How can I try FLUX 3?+
Use the official Black Forest Labs Early Access path. On this page, you can create clips with Kling 3.0 while reviewing verified information about the new release.
Compare Related AI Video Models
Explore working video generators while FLUX access rolls out.
Create an AI Video Now
Explore the official FLUX 3 direction, then use SoraVideo's clearly labeled Kling 3.0 generator for a working video workflow today.
FLUX and FLUX 3 are products of Black Forest Labs. SoraVideo is independent and is not affiliated with Black Forest Labs. Information on this page is based on the official July 23, 2026 Early Access announcement. The embedded generator currently runs Kling 3.0.