Updated: Oct 2, 2026
base model
This page hosts two FLUX 3 releases from Black Forest Labs: FLUX 3 Image for image generation and editing, and FLUX 3 Video for video with native audio. Pick the version above.
FLUX 3 Image
FLUX 3 Image is the image release of BFL's FLUX 3 multimodal architecture. BFL says it can "synthesize and edit images in a wide variety of styles, aspect ratios, and resolutions," with a significant improvement on complex prompts and accurate text rendering in multiple languages. Like FLUX 3 Video, it runs through an API under private weights, so this version has no files to download and does not take LoRAs.
Text to Image - prompt only, at 1K, 2K or 4K, in aspect ratios from 21:9 to 1:2.
Image Editing - edit with up to ten ordered reference images.
Multilingual text - legible typography in multiple languages, straight from the prompt.
FLUX 3 Video
FLUX 3 Video is Black Forest Labs' video and audio generation model, and the first release built on their FLUX 3 multimodal architecture. It generates up to 20 seconds of video in a single pass, with a synchronized audio track produced by the same model rather than dubbed on afterwards.
Built by Black Forest Labs and originally announced on their blog. All credit for the model goes to the BFL team. FLUX 3 Video runs through BFL's API under private weights, so there is nothing to download here - Civitai is hosting the generation front end so creators can run it on-site. Head to bfl.ai for API access, updates, and to support the work directly.
One Model, Video and Audio Together
BFL describes FLUX 3 as a single multimodal backbone trained jointly on images, video and audio using their Self-Flow approach, rather than a video model with an audio stage bolted on afterwards. In practice that means sound effects line up with what is happening on screen, and the model produces multilingual dialogue instead of silent footage you have to score yourself.
What Is Mirrored Here
This page covers the FLUX 3 Image and FLUX 3 Video releases. BFL has also announced FLUX 3 Dev (open-weight access to the multimodal backbone) and FLUX-mimic for action prediction. Neither is on Civitai yet, and FLUX 3 Dev is the one to watch if you want weights you can train against.
Supported Workflows
Three generation modes are wired up on-site:
Text to Video - prompt only.
Image to Video - your image becomes the first frame.
First/Last Frame - supply a start and an end image, and the model fills in the motion between them.
BFL's API also exposes keyframe-to-video, video extension and video-to-video. Those are not surfaced in the Civitai generator yet.
Draft Mode
Draft renders a fast, cheap 720p pass so you can iterate on a prompt before committing to a full render. Final output runs at 720p or 1080p, with durations up to 20 seconds and aspect ratios from 21:9 through 9:16.
How It Compares
In BFL's own preliminary preference evaluations, run at 720p on 10 second clips, FLUX 3 Video was preferred over Runway Gen-4.5 77% of the time, Luma Ray 3.2 93%, Grok Imagine Video 69%, Kling v3 Pro 60%, and both Seedance 2.0 and Gemini Omni Flash 52%. BFL singles out facial expressions, tying sounds to physical events, and multilingual output as particular strengths. These are the lab's own numbers rather than independent benchmarks, so treat them as a starting point and judge the output yourself.
Links
Announcement: bfl.ai/blog/flux-3
Model and early access: bfl.ai/models/flux-3
Black Forest Labs: bfl.ai