New accounts get 100 free credits: one 5-second 720p video.
A 10-second video in about 12 seconds
FLUX 3 Max is a fast AI video model from fal and Black Forest Labs: a prompt or a picture becomes a 5–20 s video with sound.
Average render times at 720p from fal’s benchmark; Seedance 2.5 took 277.8 s for the same clip.
Real clips made with the model. Open any prompt in the editor, or copy it and make it your own.
Static wide shot at golden hour on a desert ranch with red mesas on the horizon. A cowboy sits on a chestnut horse beside a wooden corral fence, watching the sky. A white retro rocket descends on a column of fire and lands upright next to an old wooden cabin, blasting up a wave of red dust; the horse shifts but stays calm. Roaring engine that fades to wind and settling dust.
Prompt written by us from the clip.
Cinematic night shot with 35mm film grain over a Times Square–style avenue. The camera starts on a dark rooftop with water towers, then dives over the edge toward the glowing billboards and yellow taxis below. A young woman in a pink-and-white maid dress swings between the skyscrapers on a glowing pink neon rope like a superhero, her skirt flaring, and the shot ends in a close-up as she laughs into the camera. City traffic, rushing wind and a playful whoosh.
Prompt written by us from the clip.
Anime illustration with a soft painterly, grainy texture: a girl with very long teal twin tails, a black cape over a white dress and a small red ribbon runs and spins through a sunlit meadow of red roses and blue wildflowers at the edge of a forest. Petals swirl in the wind, a rose in the foreground sweeps past the lens, and her hair streams behind her. Gentle wind, rustling leaves and a soft piano melody.
Prompt written by us from the clip.
A red paper boat drifting on a calm pond at sunrise, gentle ripples, soft birdsong
In the same preliminary blind test FLUX 3 Max ties Seedance 2.5 on image-to-video quality, and in fal’s speed benchmark it finishes a 10-second clip about 22 times sooner.
Image-to-video Elo, Artificial Analysis arena (preliminary)
Seconds to render a 10-second 720p clip, fal’s benchmark
Figures as published by fal; the arena’s confidence intervals overlap. Neither is our measurement.[fal]
FLUX 3 Max is used to make short video clips with sound from a written idea or a picture: social posts, product shots, animated illustrations, storyboards and B-roll. Because a clip comes back in seconds, it suits trying many versions of a shot and keeping the best one.
5 to 20 second vertical or square videos for TikTok, Reels and Shorts, with music, ambience or dialogue generated in the same pass.
Animate a product picture into an orbit, a reveal or a splash. The photo stays the first frame, so the product looks like itself.
Bring an anime frame, a painting or a poster to life while keeping its drawn style for the whole clip.
Turn a scene description into a moving shot in seconds to test framing, camera moves and timing before a shoot or an edit.
Fill gaps in an edit with establishing shots, weather, crowds or details that were never filmed.
With First + last frame, give the start and end pictures and let the model create the motion between them.
FLUX 3 Max renders a 10-second 720p clip in 12.45 seconds on average in fal’s benchmark, about ten times faster than FLUX 3 and 22 times faster than Seedance 2.5. Of the models fal compared, only H3 Max, fal’s own speed-tuned version of MiniMax H3 (7.55 seconds), is quicker.
| Model | Image-to-video Elo | Time for a 10 s clip |
|---|---|---|
| FLUX 3 Max | 1,139 | 12.45 s |
| Gemini Omni Flash 1.1 | 1,134 | 47.37 s |
| MiniMax H3 (768p) | 1,141 | 83.26 s |
| FLUX 3 | 1,089 | 126.44 s |
| Wan 3.0 | 1,127 | 180.16 s |
| Seedance 2.5 | 1,139 | 277.80 s |
Elo from Artificial Analysis’ preliminary image-to-video arena (AA-Video-I2V v1.1); times are fal’s averages for a 10-second, 16:9 clip at 720p (768p for MiniMax H3).[fal]
Prices and pack sizes stay the same; each second of video simply takes half the credits, so every pack and plan makes twice as much video.
720p
20
credits per second until Nov 7, then 40
1080p
34
credits per second until Nov 7, then 68
Credit packs
from $11.90
one time, never expire
New accounts get 100 free credits for a first 5-second video, no credit card.
FLUX 3 Max was post-trained by fal Research in collaboration with Black Forest Labs on FLUX 3, Black Forest Labs’ first video model, and released on fal on October 19, 2026.[PR Newswire][fal]
fal Research tuned Black Forest Labs’ first video model for sharper detail, steadier motion and closer prompt following.[PR Newswire]
Inference work cut a 10-second clip from 126.44 s to 12.45 s and lifted its arena score from 1,089 to 1,139.[fal]
fal released it on October 19, 2026 and says post-training continues, so expect it to change.[PR Newswire]
| Type | AI video model (text to video and image to video), with generated sound |
|---|---|
| Made by | fal Research with Black Forest Labs, post-trained on FLUX 3[PR Newswire] |
| Released | October 19, 2026, on fal[PR Newswire] |
| Length | Any whole number of seconds from 5 to 20 |
| Resolution | 720p render (1280 × 704 at 16:9); 1080p is upscaled[VentureBeat] |
| Aspect ratio | Auto, 16:9, 9:16, 1:1, 4:3, 3:4, 21:9, 2:1 |
| File | MP4, H.264 video and AAC audio, 24 frames per second |
| Price on Flux3Max | 40 credits per second at 720p, 68 at 1080p (half until November 7, 2026) |
Four steps and about a minute, all in the browser.
Go to flux3max.net and sign in with Google. Nothing to install and no API key; new accounts get free credits for a first 5-second 720p video.
Text to video starts from a prompt. Image to video animates a picture you upload, which becomes the first frame. In the prompt, name the subject, the one main action, the camera move, the style and the sounds.
Any length from 5 to 20 seconds, 720p or an upscaled 1080p, an aspect ratio from 21:9 to 9:16, and generated sound or silence. The credit price updates as you choose.
FLUX 3 Max renders the clip, usually in well under a minute for a short 720p video. The MP4 appears next to the editor and in My videos for 7 days. You are only charged when it is delivered.
You can try FLUX 3 Max online at Flux3Max (flux3max.net), a web app that runs the model in your browser with no download and no API key; new accounts get free credits for one 5-second 720p video. Developers can call the same model through an API on fal, where it launched, or apimodels.app.
Creators who want to make videos now
Web app: sign in with Google, no API key, first 5-second 720p clip free, then pay per second with credits.
Developers using several models
One API key for FLUX 3 Max and other image and video models, billed per second.
FLUX 3 Max is an AI video model that turns a text prompt or a still image into a 5 to 20 second video with synchronized sound. fal Research post-trained it with Black Forest Labs on FLUX 3, Black Forest Labs’ first video model, and released it on October 19, 2026. It is built for speed: fal measures about 12 seconds to render a 10-second clip.
Not entirely: FLUX 3 Max is paid per second of video wherever you use it, but you can try it free. On Flux3Max, new accounts get 100 free credits, enough for one 5-second 720p video, with Google sign-in and no credit card. After that you buy a credit pack once, from $11.90 for 1,000 credits, with no subscription and no expiry; a video costs 40 credits per second at 720p. Until November 7, 2026 (00:00 UTC), every video costs half the credits: 20 credits per second at 720p instead of 40, and 34 at 1080p instead of 68, so every pack and plan makes twice as much video; after that the regular rates return.
FLUX 3 Max is level with the best image-to-video models in a preliminary blind test and far faster than most: in Artificial Analysis’ preliminary arena it scored about the same as Seedance 2.5 and just behind MiniMax H3, while fal’s benchmark puts its render time for a 10-second clip at 12.45 seconds against 83 to 278 seconds for those models.
FLUX 3 Max is the wrong pick if you need 4K or native 1080p (it renders at 720p and 1080p is an upscale), a single take longer than 20 seconds (Seedance 2.5 goes to 30), or reference videos, audio or editing of an existing clip. It is also a first iteration: its arena scores are preliminary and fal is still training it.
fal Research post-trained FLUX 3 Max in collaboration with Black Forest Labs, on top of Black Forest Labs’ FLUX 3, and released it on fal on October 19, 2026. Flux3Max is an independent web app that runs it and is not affiliated with either company.
fal released FLUX 3 Max on October 19, 2026, as a first iteration it is still post-training, so expect the model to keep changing. It builds on FLUX 3, Black Forest Labs’ first video model.
Yes. FLUX 3 Max is an AI video generator: it makes 5 to 20 second videos with sound from a prompt or an image, not still images. The FLUX name is best known from Black Forest Labs’ image models; FLUX 3 is the company’s first video model, and FLUX 3 Max is the fast version of it. To get a still, generate an image with an image model and animate it here.
No. FLUX 3 Max can start a video from your image, which becomes the first frame, but it does not change the image itself. Edit the picture in an image editor or image model first, then animate the result with image to video.
Yes. A soundtrack is generated together with the picture unless you choose Silent. Name the sounds you want in the prompt: quiet, slow scenes are the most likely to come back with little audible sound.
Yes. Upload a JPG, PNG or WebP image, describe the motion, and FLUX 3 Max animates it, keeping your image as the first frame. Image to video is the mode it ranks best in. With First + last frame you add a second image for where the clip should end.
Yes. Write a prompt and FLUX 3 Max creates the whole shot, with sound, from nothing else. Describe one main action, the camera move, the style and the sounds you want; with the Auto aspect ratio a text-to-video clip is 16:9.
Yes, and only online: FLUX 3 Max runs on cloud GPUs and its weights are not published. On Flux3Max you use it in any modern browser, on a computer or a phone.
Any whole number of seconds from 5 to 20. For a longer piece, generate several shots and cut them together; the last frame of one shot can be the first frame of the next in image-to-video mode.
Yes, but 1080p is upscaled: the model renders at 720p and the 1080p option enlarges that output. It costs 68 credits per second instead of 40. For social clips viewed on phones, 720p is usually enough.
Yes. Choose the 9:16 aspect ratio and a length from 5 to 20 seconds, and the clip comes back as a vertical MP4 with sound that you can post as it is. 720p is usually enough for phone screens. For a longer post, generate several shots and cut them together in your editing app.
JPG, PNG or WebP files up to 10 MB and 4096 × 4096 pixels. Only use images you own or have permission to use.
40 credits per second at 720p and 68 at 1080p: a 5-second 720p clip is 200 credits and a 10-second 1080p clip is 680. Credits come in one-time packs from $11.90 for 1,000, which never expire; monthly plans from $19.90 cost less per credit if you make videos every month. Until November 7, 2026 (00:00 UTC), every video costs half the credits: 20 credits per second at 720p instead of 40, and 34 at 1080p instead of 68, so every pack and plan makes twice as much video; after that the regular rates return.
No. You sign in with Google and the free welcome credits are added to your account. The free welcome credits are for accounts created in the US, Canada, the UK, the EU/EEA, Switzerland, Australia, New Zealand, Japan, South Korea, Singapore, Taiwan, Hong Kong, Israel and the UAE, up to two accounts per network.
No. Credits are held when a video starts and only taken when it is delivered. If generation fails or the safety check refuses the request, the credits go back automatically.
Generated videos and uploaded images are deleted after 7 days. They are listed in My videos until then; download any you want to keep.
On image-to-video quality they are level in Artificial Analysis’ preliminary arena. FLUX 3 Max is about 22 times faster in fal’s benchmark (12.45 versus 277.80 seconds for a 10-second clip), while Seedance 2.5 makes takes of up to 30 seconds and can edit or extend video from reference files. Choose FLUX 3 Max for speed and Seedance 2.5 for length and references.
They are level on quality in Artificial Analysis’ blind tests: MiniMax H3 leads image to video by about two Elo points and FLUX 3 Max leads text to video by about one, both well inside the margin of error. FLUX 3 Max renders a 10-second clip in 12.45 seconds against 83.26 for H3 in fal’s benchmark and makes takes up to 20 seconds; H3 works from reference images, video and audio, outputs 2K and edits existing video.
In fal’s speed benchmark, FLUX 3 Max renders a 10-second 720p clip with sound in 12.45 seconds, faster than every other model in its table, from Gemini Omni Flash 1.1 (47.37 s) to Seedance 2.5 (277.80 s). Only H3 Max, fal’s speed-tuned MiniMax H3, was quicker at 7.55 seconds, and it scores lower in Artificial Analysis’ image-to-video arena. These are fal’s measurements, not ours.
They do different jobs. Midjourney and GPT Image are best known as image generators, while FLUX 3 Max makes videos with sound. A common workflow uses both: make the still with an image model, then animate it with FLUX 3 Max image to video.
Yes, through API providers that sell it per second of video: fal, where it launched, apimodels.app and Krea. Black Forest Labs’ own API offers the base FLUX 3 video model but not FLUX 3 Max. Flux3Max itself is a web app for making videos without code and has no API.
Still unsure? Email support@flux3max.net or read the About page.
Figures from model makers are their own measurements unless stated.