FLUX 3 Max Image to Video

FLUX 3 Max Image to Video

What is FLUX 3 Max Image to Video?

FLUX 3 Max Image to Video turns a single still into a moving clip with sound. Give it a first frame — a PNG, JPEG or WebP — plus a prompt describing how the shot should move, and it animates the image into up to 20 seconds of 720p or 1080p video with native, synchronized audio. It is a frontier image-to-video model, a post-trained version of Black Forest Labs' FLUX 3 multimodal foundation, and it shares the same model and pipeline as FLUX 3 Max Text to Video.

The model is built for speed at frontier quality: a 10-second clip returns in roughly 12 seconds, so the take you iterate on is the take you ship. On Artificial Analysis's preliminary image-to-video arena it sits tied for the top score, with faces and fine materials that hold up in close-up.

Key Features

• •Animate from a first frame, with an optional last frame for controlled first-to-last transitions.
• •Native synchronized audio rendered in the same pass, so an impact sound lands on the right frame.
• •Motion that keeps its shape: fast subjects hold the form they had in the first frame instead of smearing or morphing.
• •Duration 5 to 20 seconds; resolution 720p or 1080p; aspect ratios from 21:9 to 9:16, or auto to follow your image.
• •Reasoning level low or high, and a seed for reproducible runs.

Best Use Cases

Bring product shots, portraits and concept frames to life for social and ad clips; add motion and sound to a storyboard frame for rapid previs; animate performance, sport and machinery stills where rigid shapes must stay consistent; and build first-to-last transitions by supplying both ends of a shot. In testing, a still of taiko drummers with raised sticks animated cleanly into the downbeat strike — arms and bachi held their shape through the motion blur while synchronized drum booms landed on each visible hit.

Prompt Tips and Output Quality

Compose the first frame at the top of an action so the motion has somewhere to go. In the prompt, name one verb-driven action, a camera move motivated by that action, and one line of sound. Keep audio on for any scene with a clear sound event, and use high reasoning for busy, multi-subject motion. Output is clean h264 video with an AAC audio track at the resolution you choose.

How to Use FLUX 3 Max Image to Video

FLUX 3 Max Image to Video animates a still you provide into a short clip with native sound. Upload a first frame, write a prompt that describes the motion, and the model brings the image to life — keeping the subject's shape from that first frame while a synchronized audio track lands each sound on the right moment.

Start with the first frame

The image is your first frame (PNG, JPEG or WebP) and the clip begins exactly there, so composition matters. Frame the subject at the top of an action — a raised stick, a lifted foot, a tilted pour — so the model has an obvious motion to carry forward. A clean, well-lit frame with a clear subject animates more predictably than a cluttered one.

Write the prompt for motion

Describe what moves, not just what is in the frame. Name one verb-driven action, a camera move motivated by that action (a lateral dolly, a slow push past the subject), and one line of sound. Specific, physical language reads better than mood words. If you want a defined ending, add a last_frame_image and the model interpolates a controlled first-to-last transition.

Recommended Settings

• •generate_audio: keep true for any scene with a clear sound event — native synced audio is the model's signature.
• •reasoning_level: use high for busy, multi-subject or fast motion; low is faster for simple animations.
• •duration / resolution: 5 seconds at 720p for drafts and iteration; raise duration and move to 1080p for finals. Both are billed per second of output.
• •aspect_ratio: auto follows your first frame; set it explicitly to match the frame's shape.
• •seed: fix a value to reproduce a take; use -1 for variety.

In testing, a first frame of three taiko drummers with raised bachi animated into the downbeat at 5s / 720p / 16:9 with audio on and high reasoning (seed 824610357): the sticks and arms held their shape through fast motion blur and the drum booms stayed locked to each strike — a clean, reproducible result that shows what the model does best.

Examples to try

Each example below ran end to end at the settings on its card — 5 seconds, 720p, 16:9, audio on, high reasoning — animating the first frame shown, with all motion, the camera move and the sound coming from the prompt.

• •Animate a busy still with synced audio (hero). Prompt: "The three drummers drive their raised bachi down into the drums in unison, lunging low on the downbeat, arms snapping through the strike and sweat flicking off them; the red paper lantern sways behind them. A low lateral dolly tracks past the drum skins at the moment of impact. Audio: thunderous synchronized taiko booms locked to each strike, sharp rim clacks, a shouted 'Sore!', a crowd murmuring in the dark." Arms and sticks hold their shape through the blur and each boom lands on its strike. Tip: frame the first still at the top of the action.
• •Fast multi-subject motion that keeps its shape. Prompt: "The four racehorses surge around the dirt bend in a tight pack, legs scissoring and clods of earth flung up behind the hooves, the jockeys rising and dropping in their bright silks; the camera tracks alongside at rail height, level with the leader's shoulder. Audio: a drumming gallop of hooves on packed dirt, short shouted calls from the jockeys, a distant grandstand roar." Legs, jockeys and flung dirt stay legible through the motion blur. Tip: shoot the first frame at full stride and keep reasoning_level high.
• •Faces and fine materials in close-up. Prompt: "The watchmaker lowers the hair-thin tweezers and seats the tiny gold gear into the open movement, his loupe eye narrowing as he holds his breath; a slow push-in stays on his hands and the brass wheels. Audio: a faint metallic click as the gear beds in, the soft tick of the movement starting, quiet room tone." Skin, the brass loupe and the watch wheels stay sharp as he moves. Tip: start from a well-lit close-up and move to 1080p for a final.

Also built for prompts like "a potter's wet hands pull a tall vase up on a spinning wheel" — or supply a last frame to morph one shot into another, such as "a closed red rosebud opens into full bloom."

Does it generate audio?

Yes — native synchronized audio, on by default. Turn it off for a silent clip.

How long can a clip be?

Up to 20 seconds in a single generation, at 720p or 1080p.

What image formats can I use as the first frame?

PNG, JPEG or WebP.

Can I control the ending?

Yes — supply a last frame and the model interpolates a controlled first-to-last transition.

Who makes it?

It is a post-trained version of Black Forest Labs' FLUX 3 multimodal model.

Is it really tied for #1?

It is tied within the confidence interval on a preliminary, company-run arena — frontier quality, not a settled ranking.