Best photos for AI breaking videos: an image-to-video checklist

The fastest way to improve an image to breaking video AI result happens before you press generate: picking the right photos. The model has to keep a person's face, hairstyle, outfit, and body shape stable while they drop into footwork, spin through power moves, and hit a freeze — so give it the clearest reference you can.

Use head-to-shoes framing

A half-body portrait may look great as a profile picture, but it gives the model no reference for legs and shoes — and breaking is a full-body dance. Choose photos where the whole person is visible, with a little empty space around the head and feet. A hand or shoe touching the image edge is more likely to turn inconsistent once the dancer starts moving.

Front-facing and well-lit beats angled and moody

The model reads identity from the face and silhouette. Photos that work best are:

• front-facing, with the face unobstructed (no sunglasses, masks, or heavy shadows);
• evenly lit — soft daylight or plain indoor light rather than harsh flash, colored filters, or strong beauty effects;
• sharp at full size. Zoom into the face and shoes before uploading: blur in, blur out.

One person per photo

A clean single-person photo preserves more detail than a group shot the model has to interpret. Group photos invite borrowed hair, hands, or clothing from the person standing next to your subject. If you want two people in the routine, use two separate photos — the second slot exists for exactly that, and clearly different outfits help the model keep the two identities apart.

Choose simple, visible outfits

The generator uses each photo's clothing as a reference. Both fitted and loose clothes work, but the full outfit should be visible and contrast with the background. Very dark clothes against a dark wall, long fabric hiding the shoes, or large objects held in front of the body make the moves harder to infer.

Only upload photos you may use

Upload yourself, or people who have clearly agreed to appear in your video. Do not upload celebrity photos, deceptive edits, or images of children without a parent or guardian's permission. Photos that fail content review stop the task — though the credits are always returned automatically.

A quick upload checklist

• One consenting person per photo, two photos at most
• Head-to-shoes framing with space around the body
• Face, hands, and shoes unobstructed and sharp
• Simple outfit that contrasts with the background
• JPG, PNG, or WebP, no larger than 10MB each

If a result has an unstable face or unusual hands, improve these inputs first rather than writing a longer prompt — better references solve more problems than extra adjectives.