Flux 3

How Flux 3 works

FLUX 3 is Black Forest Labs’ frontier video model. One cook bar, four capabilities. The number of stills you attach is the only switch — there is no separate mode picker.

Outputs are .mp4, at 720p or 1080p, 5 to 20 seconds, with native audio on by default. Stills should be PNG, JPEG, or WebP. Start videos are not supported.

Capabilities

Same prompt, ratio, resolution, duration, and audio controls for every run.

• text-to-video — 0 stills. The prompt becomes the whole shot.
• image-to-video — 1 still. That image is the first frame; the prompt describes the motion and sound.
• first-last-frame-to-video — 2 stills. Flux 3 interpolates a smooth transition from Start to End.
• keyframes-to-video — 3 to 10 stills. Each still is pinned along the timeline; Flux 3 generates the motion between them.

How to use

Everything happens in the cook bar at the bottom of the page.

• 1. Prompt — describe the motion, camera, and sound. Spoken lines in the prompt can become dialogue when audio is on.
• 2. Stills — skip them, or tap the image button and upload / pick from your library (up to 10). Order is timeline order. Thumbs are labeled Start / End (and 2, 3… in between). Remove and re-add to reorder.
• 3. Settings — ratio, 720p or 1080p, duration 5–20 seconds, audio on or off.
• 4. Cook — the price in the button is resolution × duration. Closing the tab after cook starts is fine; we email you when the clip is ready.

Text to video

Generate a clip from a written description — motion, composition, and scene, with optional native audio. No input image.

• How — leave the image tray empty. Write the prompt, set ratio / resolution / duration, cook.
• Ratio — auto, or 21:9, 2:1, 16:9, 4:3, 1:1, 3:4, 9:16.
• Best for original scenes, period looks, music videos, and action from text alone.

Image to video

Animate a single still into coherent motion. The image is the first frame; the prompt extends it into the rest of the clip.

• How — attach exactly one PNG, JPEG, or WebP. Describe how it should move and sound, then cook.
• Ratio — auto lets the model read the still, or set it yourself.
• Best for bringing a hero still to life — cinematic push-ins, atmospheric scenes, and natural motion from one frame.

First last frame to video

Generate the video between a defined start and end frame. Flux 3 interpolates a smooth, coherent transition from the first image to the last. An explicit duration is required so the end frame can be placed.

• How — attach exactly two stills. The first thumb is Start (opening frame), the second is End (closing frame). Describe the transition in the prompt, then cook.
• Duration — pick 5–20 seconds. The end still is pinned at the last frame of that length.
• Best for before/after, day-to-night, matching two boards into one shot, or a clean start-to-end morph. Keep both stills similar in framing for a cleaner transition.

Keyframes to video

Build video from a sequence of keyframes. Flux 3 generates the motion between each anchor so you control how a shot progresses. First still is pinned at frame 0, last at duration × 24 fps; the rest are spaced evenly, with unique indexes (up to 10 stills).

• How — attach 3 to 10 stills in timeline order. Thumbs read Start, 2, 3… End. Describe the motion between anchors, then cook. Adding a third still is what switches from first–last-frame to keyframes.
• Duration — pick 5–20 seconds. Positions are spaced across that length. Example: 5s with three stills uses frames 0, 60, and 120.
• Best for product sequences, multi-beat shots, and matching several storyboards into one continuous clip.

Settings

prompt

Required. Describe the motion, camera, and sound you want.

images

None for text-to-video, one to animate, two for first-to-last frame, or 3–10 for keyframes. PNG, JPEG, or WebP.

aspect_ratio

auto, or one of 21:9, 2:1, 16:9, 4:3, 1:1, 3:4, 9:16. auto lets the model choose.

resolution

720p (default) or 1080p.

duration

5 to 20 seconds. Two stills place the end frame at this length. 3–10 stills are spaced across it.

generate_audio

on by default. Turn it off for a silent clip.

About

FLUX 3 is built on Black Forest Labs’ Self-Flow architecture, a single multimodal flow-matching model trained across images, video, and audio.

FAQ

What can Flux 3 generate on Soup.video?

Flux 3 has four capabilities on one form. Zero stills is text-to-video. One still is image-to-video. Two stills interpolate first-last-frame-to-video. Three to ten stills are keyframes-to-video. Native audio is optional and on by default. Start videos are not supported.

How do I use Flux 3?

Write a prompt describing motion, camera, and sound. Attach stills in timeline order or skip them. Set aspect ratio, 720p or 1080p, duration from 5 to 20 seconds, and audio. Cook. The number of stills picks the mode. Closing the tab after cook starts is fine; Soup.video emails you when the clip is ready.

How does text-to-video work in Flux 3?

Leave the image tray empty. The prompt becomes the whole shot. Set ratio, resolution, and duration, then cook. No input image is needed.

How does image-to-video work in Flux 3?

Attach one PNG, JPEG, or WebP. That still is the first frame. The prompt describes how it should move and sound.

How does first-last-frame-to-video work in Flux 3?

Attach exactly two stills. The first is the opening frame and the second is the closing frame. Flux 3 interpolates a smooth transition between them. Duration from 5 to 20 seconds is required so the end frame can be placed.

How does keyframes-to-video work in Flux 3?

Attach three to ten stills in timeline order. First is pinned at the start, last at the end, and the rest are spaced across the duration at 24 frames per second. Adding a third still switches from first-last-frame to keyframes.

What resolutions, durations, and prices does Flux 3 support?

Flux 3 outputs MP4 at 720p or 1080p, from 5 to 20 seconds, with optional native audio. Pricing is per second for all four capabilities: ✦ 25/s at 720p and ✦ 40/s at 1080p. A 5-second 720p clip is ✦ 125.