How to Make a Video from a Photo with AI in 2026: Step-by-Step

Video from a photo: what it is and why it matters

Making a video from a photo with AI means uploading a still and getting a short clip where a person, hair, background, or camera moves. Technically this is image-to-video: the model uses your image as frame one and invents the next seconds.

Common uses: animate a portrait for Reels, bring a product card to life, make a living greeting, or ship a teaser without a shoot. On Gensta.ai, image-to-video works with MiniMax, Sora 2 / Pro, Veo 3.1, Wan, and Seedance 2.0. Separately, Kling Motion Control transfers exact moves from a reference video onto your photo.

This is not an 'undress photo' tool and not a deepfake of someone else: photos without consent and porn content are banned.

Prepare the photo: checklist before you generate

A good source does half the quality work.

Works well: sharp photo from 720×720 (ideally 1024×1024+), subject well lit, one or two main people, natural pose. Fails often: messenger compression, watermarks across the frame, very dark or blown-out shots, huge group photos.

For old film photos, restore first, then animate — see the old-photo animation guide. For ready scenarios without manual prompting, open the Photo Animation presets collection.

Step-by-step: make a video from a photo on Gensta.ai

Step 1. Open Create → Video tab. Step 2. Upload the photo as the start frame. Step 3. Write a motion prompt (not an appearance prompt): what moves, how the camera moves, what tempo. Step 4. Pick a model: fast tests — MiniMax 2.3 Fast; people/emotion — Sora 2; camera + optional audio — Veo 3.1; art — Wan 2.5; exact dance/gesture — Kling Motion. Step 5. Use 6–10 seconds on the first run and hit Create.

If the result is weak, don't change model and prompt at once. First simplify to one verb, then try a second model.

Which model to choose for video-from-photo

• MiniMax 2.3 Fast — fastest start for image-to-video specifically (no text-to-video on Fast), 6/10s, 768p/1080p. • MiniMax 2.3 — same class with text-to-video available when needed. • Sora 2 / Sora 2 Pro — strong physics and faces; Pro up to 1080p. • Veo 3.1 / Fast — camera moves and optional audio. • Wan 2.5 — art, animation, stylization. • Seedance 2.0 — references and longer clips (up to 15s, 480p/720p). • Kling Motion Control — not 'just animate', but copy moves from your reference video.

Rule of thumb: draft on Fast → finalize on Sora/Veo/Seedance. For dances, go Motion first.

8 motion prompts + common mistakes

Prompts (adapt to your photo): 1) "slowly smiles, blinks once, light hair movement, face stays stable". 2) "turns head left, looks at camera, natural expression". 3) "camera slowly pushes in on the face, soft light, cinematic". 4) "hair and clothing fabric move in a breeze, background slightly alive". 5) "person steps toward camera, smooth motion, vertical frame". 6) "product on a table, camera slowly arcs around it, studio light". 7) "foliage and highlights move in the background, portrait almost still, calm rhythm". 8) "mild surprise: raises eyebrows and smiles slightly, no cartoon grimace".

Mistakes: too many actions in one prompt; aggressive motion on a tight close-up; bad source photo; expecting MiniMax Fast to make a clip from text with no image.

FAQ: AI video from a photo

How many seconds do I get? Depends on the model: often 4–12 seconds; Seedance/Wan 2.7 up to 15. Can I get audio immediately? Yes — Sora and Wan usually include it; Veo lets you toggle audio. How is this different from 'animate a photo'? Same intent family; the animate-photo guide focuses more on portraits and presets; this one covers i2v methods and model choice. Can I animate someone else's photo without consent? No. Banned on Gensta.ai.

Next: open MiniMax Fast with your photo or the Photo Animation collection and make the first clip in a couple of minutes.