FLUX 3 Max · FLUX 3
Turn a prompt or a still into a clip of up to 20 seconds, with dialogue and sound made in the same pass. FLUX 3 Max, the faster post-trained edition, can also land on an end frame you choose; FLUX 3 rewrites a clip you already have. The Generate button shows the exact credits before you start.
FLUX 3 Max is fal Research's post-trained edition of FLUX 3, built with Black Forest Labs and released on October 19, 2026. The extra training aimed at closer prompt following, steadier motion and far shorter waits. fal's own timings and blind-test scores are in the comparison table below.
Speed is the headline. A take comes back quickly enough that you can rewrite the prompt and run the shot again instead of settling for the first result. On Oeneo AI the wait also includes upload and queue time, so read fal's timing as the model's pace rather than a promise.
Write the beats of a scene one after another, such as an arrival, a handoff and a reaction, and FLUX 3 Max plays them in that order inside one shot. Short ad scenes and storyboard beats can come from a single prompt. Name who does what and keep each beat to one action.
In image to video, FLUX 3 Max takes an optional second image as the end frame and works out the motion between the two. A product shot can land on its packshot, or a scene can end on the exact frame your storyboard needs. FLUX 3 starts from a first frame only on Oeneo AI.
Both models make 5–20 second clips at 720p / 1080p, with optional sound and the same framing choices. The rows below are where they differ. Scores and timings are the figures fal published from the preliminary Artificial Analysis image-to-video arena.
| Specification | FLUX 3 Max | FLUX 3 |
|---|---|---|
| Made by | fal Research, post-trained on FLUX 3 | Black Forest Labs |
| Released | October 19, 2026 | July 23, 2026 |
| Image-to-video blind test | 1,139 EloTied with the leaders, per fal | 1,089 Elo |
| Time for a 10-second clip | 12.45 s | 126.44 s |
| Image to video | First frame + optional end frame | First frame only |
| Video editing | No | 720p, sources up to 15 s |
| Pick it for | Fast takes, ordered beats, set endings | Reworking a clip you already have |
| Try it | Use FLUX 3 Max | Use FLUX 3 |
Brief the shot the way a director would: one clear event, what the camera does and what we hear. The same advice works for FLUX 3 Max and FLUX 3.
Build the shot around one clear event
A marble knocks over a row of dominoes; a cyclist rides into a tunnel. Motion that passes from one thing to the next reads cleanly.
Say what the camera does
Slow dolly in, low-angle close-up, side-on tracking shot. Leave it out and the model chooses for you.
Name the sounds and quote the lines
"A single bell strike", "rain on a tin roof", or a quoted line with the character who says it.
List the beats in the order they happen
Arrives, hands over the bag, laughs. With a first frame, describe only what changes after it.
Many bodies moving in sync
A crowd crossing one spot or a full-body routine is harder to get right; split it into separate shots.
A prompt that fights the first frame
If the photo shows a beach at noon, asking for a city at night pulls the clip two ways at once.
A quiet scene with no sound cue
Still, slow shots can come back nearly silent unless the prompt asks for a sound.
Settings this site does not offer
Negative prompts, keyframes, Draft mode and 2K or 4K output are not available here.
Example: "A street-food cook flips noodles in a wok over roaring flames at night, close side-on shot, steam rising, sizzle and night-market chatter."
Upload a clip of up to 15 seconds, pick FLUX 3 with video to video, and describe the change. The model renders the clip again at 720p, keeping its motion, timing and framing, and changes what the prompt names. FLUX 3 Max does not edit video.
Name the object and its replacement. The knife, the camera move and the timing stay where they were, so the new version drops into the same edit without re-timing.
Edit a clip with FLUX 3Rough 3D previz can become a finished scene: the source clip supplies the camera path and pacing, and the prompt supplies the place, vehicles, light and dust.
Ask for a new palette, time of day or lighting setup. The performer moves exactly as filmed, which makes it a quick way to test looks for a music video or an ad.
A few of the looks the family covers, from anime to lettering that sits inside the scene. Each clip keeps its original sound.
A finished clip can go straight into the Oeneo AI video tools: raise its resolution, add seconds to the end, cut out the background, or use it as the motion for a photo.
Cut the subject out of any clip and export a transparent WebM or a solid colour.
Open toolFLUX 3 is Black Forest Labs' first video model: one multimodal model trained on images, video and audio together, announced on July 23, 2026. FLUX 3 Max followed on October 19, 2026, when fal Research post-trained it together with Black Forest Labs for closer prompt following and much faster generation. This page runs both.
No. FLUX 3 Max is the same FLUX 3 base with extra training and inference work by fal Research, done in partnership with Black Forest Labs; it is not a larger tier that Black Forest Labs released itself. In use, it returns takes faster and accepts an end frame, while FLUX 3 is the one that edits video.
Open the Workbench at the top of this page, which starts on FLUX 3 Max, or switch to FLUX 3. Write the shot, or add a first frame for image to video, then set the length, resolution and aspect ratio and click Generate. The button shows the exact credits first, and the finished clip downloads with its sound.
No, each clip costs credits. The price depends on the model, the length and the resolution, and the Generate button shows the exact amount before you start. New accounts get 10 credits.
No. FLUX 3 Video and FLUX 3 Max are hosted models whose weights have not been released, so they do not run in ComfyUI or on your own GPU. Black Forest Labs has said an open-weight FLUX 3 Dev is part of its rollout. On Oeneo AI both models run in the browser with nothing to install.
Any whole number of seconds from 5 to 20, in one continuous take, on both models. For something longer, add seconds to the finished clip with the Video Extender or plan the piece as several shots.
Yes. Speech, effects and ambience are generated with the picture, and sound is on by default; switch it off for a silent clip. Lines in quotes are spoken on screen, and naming the sounds you want helps most in quiet, slow scenes. Edits made with FLUX 3 always come back with sound.
FLUX 3 Max takes a first frame plus an optional end frame. FLUX 3 starts from a first frame only on Oeneo AI. Multi-keyframe shots and Draft previews are not available here.
No. Both models output 720p / 1080p on Oeneo AI. For a larger master, run the finished clip through the Video Upscaler.
Close on quality, far apart on speed. In the preliminary Artificial Analysis image-to-video arena that fal cites, FLUX 3 Max scored 1,139 Elo, level with Seedance 2.5 at 1,139 and just under MiniMax H3 at 1,141, with overlapping margins. fal timed a 10-second clip at 12.45 seconds for FLUX 3 Max, against 83.26 for MiniMax H3 and 277.8 for Seedance 2.5.
No. This page covers the FLUX 3 video models. FLUX 3 Image is Black Forest Labs' still-image model and has its own page on Oeneo AI.
Text or image to video
Describe a shot or add a still, pick a length of up to 20 seconds, and get a clip with its own soundtrack.
