Flux 3 First Last Frame to Video: the still is the contract (5s, 6s, 7s, 9cr)

Flux 3 First Last Frame to Video is image-to-video. If the label is wrong, every second is waste.

Black Forest Labs, category image-to-video, requires_image: true, 9 credits, audio: true. Description: "FLUX.3 interpolates a smooth, coherent transition between a defined start frame and end frame, with native audio — up to 1080p and 5–20 seconds." Durations: 5s through 20s, every second. Aspects: 21:9, 2:1, 16:9, 4:3, 1:1, 3:4, 9:16, auto. Qualities: 720p, 1080p. This is not a one-still "make it move" row. The contract is two stills. If either end is the wrong identity, every second of the interpolation is a morph between two mistakes, with native audio attached, at 9 credits a second.

Two stills, one interpolation

The prompt is connective tissue. It does not replace frame one or frame two. Describe the path: how light shifts, how the subject travels, how the cut should feel. Re-describing "a red bottle on a table" when the first still is already that bottle wastes the only prose this family actually needs — the in-between.

5s, 6s, 7s (the title examples) sit on a list that continues to 20s. A 5-second morph and a 20-second morph are different bills at 9 credits per second. Size the interpolation to the change. A tiny wardrobe delta does not need 20 seconds. A day-to-night sky might.

Native audio is on. Plan for a soundtrack the model will generate with the transition. If you need silence, this is a louder SKU than Kling's silent motion-control row. If you need a specific VO, this is not lipsync — it is interpolation with native audio, not a mouth pass against your track.

Image to video is the door. Create a first-and-last-frame transition is the agent. The first-last-frame glossary is the mechanic. Black Forest Labs' roster holds Flux.3 video. Flux 3 Text to Video is the no-still cousin. Flux 3 Extend Video continues a clip. This page is the two-still SKU.

If you only have one still, the label is wrong

One hero frame and a camera-move prompt is ordinary image-to-video, not first-last-frame. Two hero frames that do not share identity — different faces, different products — is a morph between strangers. The stills have to be a pair: same subject, composed so a path between them is plausible.

No listed max-resolution field; qualities go to 1080p. Do not brief 4K. 9 credits is not a 4K invoice.

Flux's family register (composition, colour, mood, photographic language) is useful here as a description of how the frame evolves, not as lens hardware this schema does not expose. "Warm key cools as the subject turns" is a path. "85mm f/1.4" is a stills leftover.

Seconds of a bad pair are still seconds

Wrong jobs:

• Text-only scene → Flux 3 Text to Video.
• One still, no end frame → a start-frame i2v row, not this.
• Need a talking head on a VO → lipsync. Native audio here is not your track.
• Need to continue a finished clip → Flux 3 Extend Video.

Every second between 5 and 20 is interpolation. If the ends are not keepers, 9 credits times duration is the cost of watching them fail in slow motion, with audio.

FAQ

Does this require an image?

Yes. requires_image is true, and the description is a start frame and an end frame. One still is the wrong label. Two stills that disagree on identity are also the wrong label.

What durations are listed?

5s through 20s, inclusive, one-second steps. 5s / 6s / 7s are the short end of that range, not a separate SKU. 9 credits meters per second.

Does it include native audio?

Yes. audio: true, and the description says so. You are buying a transition with a generated soundtrack, up to 1080p. If you need a specific VO, use a lipsync row after the picture exists.

How is this different from Pixverse Transition?

Both interpolate two stills. This Flux.3 row lists 5–20s, native audio, 9 credits, and the wide aspect set including 21:9 and 2:1. Pixverse's effects/transition family is a different catalog contract. Pick the pair of stills first, then pick the row whose duration and audio flags match the cut.