Stills animation walkthrough with AI
No keyframes, no timeline, no motion slider. Animating a still is one submission, which is exactly why it works at forty images and a deadline.
TL;DR
There are no keyframes, timeline or motion strength setting. Animating a still is one submission: image, prompt, duration, aspect ratio. For a batch, the prompt is what makes it tractable, since consistent product photography means one working sentence applies across the whole set. Decide whether the camera or the subject moves, never both, keep movement small to protect label text, and review the finished set as a group rather than clip by clip.
There are no keyframes. No timeline, no scrubber, no millisecond markers, no motion strength slider.
Animating a still is one submission: the image, a sentence about what should move, a duration and an aspect ratio. The model returns a clip. There is nothing to adjust between those two points.
That sounds restrictive until you have forty product photos and a deadline, at which point it is the reason this is fast.
The batch problem, honestly
Forty stills is forty generations. There is no queue manager, no batch upload, no way to apply one setting across a set.
What makes it tractable is that the prompt is reusable. Product photography is consistent by nature: same lighting setup, same surface, same camera distance. So one motion prompt that works on the first bottle works on the next thirty-nine.
Which changes the strategy. Do not start by animating forty photos. Start by finding the sentence, on one photo, that produces what you want. Then apply it.
Getting that sentence right is the entire job. Everything after is repetition.
Finding the sentence
Work on your most typical image, not your best one. A prompt tuned to the exceptional shot will fail on the ordinary ones.
For product stills, the motions that reliably work are small:
"Slow push in toward the bottle. Camera moves, product stays still. Reflections shift on the glass."
"Bottle rotates slowly clockwise. Camera static. Light sweeps across the label."
Three rules that matter more here than anywhere:
Decide what moves: camera or subject. Not both. A rotating product under a moving camera is where products distort, because the model is resolving two motions at once against a flat source.
Protect the label. Text on packaging is the first thing to smear. Keep the movement small and slow, and consider a negative prompt where the model accepts one — Wan 2.7 does.
Say what stays still. "Product stays still" is as useful an instruction as any description of movement.
Choosing images that survive animation
Some stills animate cleanly and some cannot.
Sharp and well lit. Any softness in the source becomes smearing in motion.
Clear separation from the background. Products blending into their backdrop blend further once things move.
Room in the direction of movement. A tightly cropped bottle has nowhere to push into.
Simple reflections. Complex specular highlights are where models invent detail that was not there.
If your set has a few images that fail repeatedly, they are usually failing for one of these reasons rather than needing different settings.
Doing the run
- Pick the typical image and find the prompt. Expect several attempts, and use a cheap model for them.
- Lock duration and aspect ratio. Vertical if it is going to social, decided now rather than cropped later.
- Run the set with the same prompt, changing only the image.
- Review as a group, not one at a time. Inconsistency across the set is more visible than any single imperfect clip.
- Regenerate the failures individually with a slightly adjusted prompt.
Step 4 is the one people skip. Forty clips reviewed individually all look fine; reviewed together, the three where the camera drifted the wrong way are obvious.
Model choice for stills
Thirty-five video models accept an image input.
Exploring the prompt — Seedance 2.0 Fast at STANDARD, or another Fast variant. Cheap attempts while you find the wording.
Running the set — the full model you settled on, once the prompt is proven.
Repeatability matters — Wan 2.7 is the only model with a seed, plus a negative prompt. If you want the same camera behaviour across a set rather than forty independent interpretations, that is the one entry that offers it.
A moving poster rather than a clip — Motion Poster is shaped for key art.
Start at image to video, and everything you generate lands in History, which matters on a run of forty.
What is not there
No keyframes or timeline. You do not place a value at frame 48 or scrub to confirm a rotation lands at 3.2 seconds.
No motion strength. 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.
No frame rate control, and no fixed canvas size you must match.
A seed on exactly one model. Wan 2.7. Instructions to "keep the seed fixed at 42" on Seedance are describing a field that model does not have.
The short version
Find the sentence on one image. Keep the motion small and decide whether the camera or the product moves. Run the set unchanged. Review as a group.
The catalog is at Models, each generation is quoted before it runs, and pack prices are on the pricing page.
Frequently Asked Questions
How do I set keyframes when animating a still?▾
You don't. There is no keyframe editor, timeline or scrubber. Animating a still is a single submission consisting of the image, a prompt describing the motion, a duration and an aspect ratio. Instructions to place a value at frame 48 or confirm a rotation at 3.2 seconds describe an interface that does not exist.
How do I animate a large batch of product photos?▾
Find one motion prompt that works on a typical image, then apply it unchanged across the set. Product photography is consistent by nature, so a sentence that works on the first bottle usually works on the rest. There is no batch upload or queue manager, so the reusable prompt is what makes volume tractable.
Why does my product distort when it moves?▾
Usually because both the camera and the subject are moving. Decide which one moves and say the other stays still, since resolving two motions at once against a flat source is where distortion appears. Keeping the movement small and slow also protects label text, which is the first thing to smear.
Which stills animate well?▾
Sharp, well lit images with clear separation from the background, room in the direction of intended movement, and simple reflections. Softness in the source becomes smearing in motion, and complex specular highlights are where models invent detail that was not there.
Can I fix a seed so the whole set matches?▾
Only on Wan 2.7, the single video model out of 37 that exposes a seed, and one of only two with a negative prompt. On every other model each generation is an independent draw. Instructions to keep a seed fixed on Seedance describe a field that model does not have.
Which model should I use?▾
Explore the prompt on a Fast variant such as Seedance 2.0 Fast at STANDARD tier, then run the set on the full model once the wording is proven. Thirty-five video models accept an image input. Motion Poster is the better tool if you want moving key art rather than a clip.
How should I review a batch?▾
As a group rather than clip by clip. Forty clips reviewed individually all look acceptable, but reviewed together the few where the camera drifted the wrong way become obvious. Inconsistency across a set is far more visible than any single imperfect clip.
