Motion Posters: How to Make One That Works
A motion poster is a still that moves a little. That constraint is the whole craft: it has to survive as a frozen image, and the movement must never become the point.
TL;DR
A motion poster must work as a still first, because that is how it is seen most of the time. Render any text at the image stage, where models handle typography well, then move only the background or only the camera, never both. Keep it to three to five seconds and loop it using first-to-last frame with the same image at both ends, so the movement returns home by construction rather than snapping back.
A motion poster is a still that moves a little. Not a video with text on it, not an animated ad — a poster that breathes.
That constraint is the whole craft. The composition has to survive as a frozen image, because most of the time a viewer sees it that way, and the movement has to be small enough that it never becomes the point.
Get that balance right and it works everywhere: a hero on a site, a story card, a pitch deck, a screen in a lobby.
Start with a poster that works still
The most common failure is generating motion first and hoping a composition emerges. It will not.
Design or generate the frame as a still image and judge it as one. If it does not hold as a poster, movement will not rescue it — it will draw attention to what is wrong.
Text especially. Video models are unreliable at typography, so any headline should be rendered at the still stage where models handle it well. GPT-Image 2.0 claims text rendering above 99% accuracy, Ideogram V4 targets best-in-class typography, and Reve 2.1 produces up to four variations with clean in-image type.
Then move as little as possible
Motion Poster is the tool. What you write into it should be almost boringly restrained.
Move the background, hold the foreground. Drifting clouds, shifting light, falling particles, a slow gradient. The subject and any text stay put.
Or move the camera, not the content. A very slow push or drift keeps everything intact because nothing is redrawn, only reframed.
Never both. Two motions in a poster is a video.
"Very slow push in. Subject and text stay fixed and sharp. Light shifts gently across the background."
Say what stays still. In poster work this matters more than describing what moves, because the default failure is everything moving slightly.
Make it loop
A poster that snaps back at the end is broken. Posters are watched repeatedly, so the loop point is visible in a way it is not for regular video.
The reliable method: first-to-last frame on Seedance 2.0 or 2.0 Fast, supplying the same image as both the start and end frame. The movement returns home by construction rather than by luck.
That single technique is the difference between a poster you can leave running and one that irritates on the second pass.
Length and framing
Three to five seconds. Long enough to register as alive, short enough to hold together and to loop without the viewer tracking the cycle.
Decide the aspect ratio before generating. Vertical for stories, square for feeds, wide for a site hero. It is set at generation and cropping afterwards throws away the composition you just spent effort on.
When it comes back wrong
There are no settings, so every fix is wording or input.
Text warped or shimmering. The video stage is redrawing it. Move the camera instead of the content, and state that the text stays fixed.
Subject drifting. The camera was unspecified. Say what it does.
Motion too dramatic. Describe something slower and smaller. "Barely perceptible drift" is a legitimate instruction.
Loop visible. Use the same image at both ends via first-to-last frame.
Neighbouring tools
A moving character rather than a moving poster — image to video, or Motion Control to transfer a performance from a driving clip.
A consistent character across several posters — thirteen models accept a reference through reference to video.
Sound — usually unnecessary. Posters are seen more than watched, and most placements are muted.
What is not real
No motion strength or particle count. Across all 37 video generation models the parameters are prompt, duration, aspect ratio and resolution, plus references, a negative prompt or first and last frames on some. Nothing exposes a particle setting or a film-grain control.
No poster templates or loop-length slider. You pick an aspect ratio and a duration.
No layers, keyframes or timeline.
No per-model effect capabilities. Claims that one model adds particles while another handles grain describe features that do not exist.
The short version
Make it work as a still. Move one thing, slowly. Loop it with the same frame at both ends. Three to five seconds.
The catalog is at Models, each generation is quoted before it runs, and pack prices are on the pricing page.
Frequently Asked Questions
What makes a good motion poster?▾
A composition that holds as a still image, with movement small enough that it never becomes the point. If the frame does not work frozen, movement will not rescue it and will instead draw attention to what is wrong.
Why does the text warp in my motion poster?▾
Because the video stage is redrawing it. Render text at the still image stage, where models handle typography well, then move the camera rather than the content and state explicitly that the text stays fixed. GPT-Image 2.0, Ideogram V4 and Reve 2.1 are the strong options for in-image type.
How do I stop the loop from snapping back?▾
Use first-to-last frame on Seedance 2.0 or 2.0 Fast and supply the same image as both the start and end frame. The movement returns home by construction. This matters more for posters than regular video, because posters are watched repeatedly and the loop point is visible.
What should actually move?▾
Either the background or the camera, never both. Drifting clouds, shifting light or falling particles while the subject and text stay put, or a very slow push that reframes everything without redrawing it. Two motions in a poster makes it a video.
How long should a motion poster be?▾
Three to five seconds. Long enough to register as alive, short enough to hold together and to loop without the viewer starting to track the cycle.
Can I set the particle count or add film grain?▾
No. Across all 37 video generation models the parameters are prompt, duration, aspect ratio and resolution, plus references, a negative prompt or first and last frames on some. There is no particle setting, no grain control and no per-model effect capabilities.
Are there poster templates or a loop-length slider?▾
No. You choose an aspect ratio and a duration as elsewhere, and there are no layers, keyframes or timeline. Decide the aspect ratio before generating, since cropping afterwards throws away the composition.
