Image to video: how to animate a still the right way

How to turn a still image into AI video: choosing a good first frame, writing motion-only prompts, and fixing the warping and drift that ruin most attempts.

October 14, 2026, 4 min read

The short answer

To animate a still image well, start from a sharp, well-composed frame and write a prompt that describes only the motion: what moves, how the camera moves, and how fast. The image already decides the look, so repeating what is in the picture wastes the prompt and can make the model change things.

Image to video is the most controllable way to make AI video. You fix the composition, the character and the product in a still, check it, and only then pay for motion. Most problems come from a weak first frame or a prompt that fights the image.

What makes a good first frame

Sharp subject, clean edges. Blur and noise in the still turn into smear and crawl in motion.
Room to move. If the camera will push in, the subject can sit a little wider. If a person will walk right, leave space on the right.
A natural mid-action pose. A person caught mid-step or mid-turn gives the model a direction. A stiff pose can stay stiff.
Simple hands and few small objects. Fingers and clutter are what warp first.
No text, logos or watermarks. Words in a still will wobble when the frame moves.
The right aspect ratio. Make the still in the same shape as the video you want, such as 9:16 for vertical. Cropping later costs resolution and framing.
Light that can plausibly move. A sunset still can animate into a gentle light shift; a flat studio image gives the model nothing to play with.

Write motion, not description

Before: "A beautiful woman with long red hair in a green dress standing in a field of sunflowers at golden hour, highly detailed, cinematic, 8K."

Why it fails: everything in that prompt is already in the picture. The model has no motion instruction, so it guesses, often with a vague zoom or a face that subtly changes because the prompt told it to render the woman again.

After: "She turns her head slowly toward the camera and smiles. A light breeze moves her hair and the sunflowers sway gently. Slow push-in. The sun flickers through the petals behind her."

Try this prompt in Create Video

The fixed prompt is all verbs and camera. It names one subject action, one environmental motion and one camera move. The look stays the same because nothing in the prompt asks for a new look.

Step by step

Make or choose the still. Use an image model or a real photo you own. In ClickPop, the image models for first frames include Seedream 5.0, Nano Banana Pro, GPT Image 2 and FLUX.2.
Check it at full size. Look at hands, eyes, edges of hair and any text. Fix or regenerate the still now; it is cheaper than fixing the video.
Decide the one thing that moves most. A person, a product, water, or the camera.
Write the motion prompt: subject action, environment motion, camera move, speed.
Pick the clip length. Shorter clips drift less. Start short, then extend if the motion holds.
Draft on a cheaper model. Seedance 2.0 Fast and Mini or Auto are fine for checking that the motion works.
Render the final on the model you want, at the resolution you need.

In Create Video, add your still as the starting image and type the motion prompt. If you are planning several shots, storyboard mode draws the stills first so you can approve every frame before anything moves.

Motion prompts you can copy

Portrait: "She blinks, takes a slow breath and glances out of the window. Her hair moves slightly. Camera holds still." Try it
Product on a table: "Steam rises slowly from the cup. Camera pushes in gently toward the logo-free side of the mug. Morning light brightens a little." Try it
Landscape: "Clouds drift left to right across the mountains, light and shadow moving over the slopes. Slow pan right."
City street: "Pedestrians walk past in both directions, a cyclist crosses the frame from left to right. Camera stays locked off."
Character action: "He pulls the hood of his jacket up, turns and walks away from camera down the alley. Camera follows slowly at shoulder height."
Food: "Melted cheese stretches as a slice of pizza is lifted out of frame. Slow motion, close-up."

How much motion to ask for

Small, motivated motion looks best. Breathing, hair moving, steam, a slow camera move. Big motion, such as a full turn, a jump or a crowd running, asks the model to invent large parts of the scene that the still never showed, and that is where faces change and bodies warp.

If you need a big change, make two stills: the start and the end. Some models accept an end frame as well as a start frame, which locks both ends of the shot. See the first and last frame guide for how to plan that pair.

Mistakes to avoid

Describing the image again. Describe only what changes.
Asking for a different style than the still. A watercolor still with a prompt for photoreal motion will fight itself.
Low-resolution stills. Upscale the image first if it is small.
Animating real people from photos without their permission. It is not allowed, and on most platforms realistic synthetic content must be disclosed.
Long clips from one still. If the motion falls apart after a few seconds, generate a shorter clip and cut to a new shot.

Questions

What should I write in an image-to-video prompt?

Write only the motion: what the subject does, what moves in the environment, and what the camera does and how fast. The still already sets the look, so describing it again is unnecessary.

Why does my image change so much when it animates?

The prompt is asking for too much motion or re-describing the scene, so the model redraws it. Ask for smaller movements, keep the clip shorter, and make sure the still itself is sharp and clean.

Which image works best for image to video?

A sharp image in the target aspect ratio, with a clear subject, simple hands, no text and some room in the frame for the planned movement.