H3Gen

MiniMax H3 Image to Video

Upload a photo as the first frame, add a last frame if you like, describe the motion and the sound, and get a clip of 4 to 30 seconds back at up to 2K.

Video
50

Image to video AI that keeps your first frame

With MiniMax H3 image to video, the photo you upload is the opening frame of the clip. Faces, logos, packaging and colours start exactly as they are, and the model adds motion, camera work and sound from your prompt. The aspect ratio follows your image, so a vertical photo gives a vertical video.

How it works

Three steps, no editing software.

  1. 01Upload the first frame

    Click the reference card and choose one PNG, JPG or WebP up to 20 MB. A sharp, well-lit image gives the cleanest motion.

  2. 02Describe the motion and the sound

    Say what moves, how the camera travels and what we hear: “she turns and laughs, slow push-in, café chatter and a soft guitar”.

  3. 03Pick a length and create

    Choose 4 to 30 seconds. Each second costs 4 credits at 480p, 6 at 768p, 8 at 1080p and 10 at 2K (2K goes up to 15 seconds; longer clips cost double per second). A failed render is refunded automatically.

What people animate from one photo

A single still is enough for these.

Product shots

Rotate a packshot, add light sweeps and a sound cue for a product page or an ad.

Portraits

Bring a portrait to life with a glance, a smile or a few spoken words.

Posters and key art

Make the characters and type on a poster move without redrawing them.

Image to video AI: what makes a good first frame

The model starts from the exact pixels you upload, so the clip can only be as clean as that frame. A sharp photo with one clear subject, even light and some space around the subject gives the camera room to move. Very small faces, heavy motion blur or busy collages tend to drift in the first second.

Think about where the motion goes before you write. A portrait framed from the chest up suits a turn of the head, a smile or a few spoken words. A product on a plain table suits a slow orbit or a light sweep. A wide landscape suits a drone push or clouds rolling past. Describing a move the frame cannot support, such as walking out of a tight close-up, usually breaks the subject.

Sound is written into the same prompt. Name the voice and what it says in quotes, then the room tone and any effects: “she says ‘new colours are in’, soft shop music, a door chime”. If you leave sound out, the model still adds audio that fits the scene.

Usually worth avoiding

  • Screenshots with small text you need to keep perfectly sharp
  • Group photos where every face has to stay recognisable
  • Images you do not own or have no permission to use

Tips for better results

  • Use the image you want the clip to open on: the first frame is kept, not reinterpreted.
  • Describe one main action per clip; add cuts only when you want them.
  • Name the sound you expect, from dialogue to room tone, or the model will pick one.
  • Text already in the image stays readable best when the camera move is gentle.

Questions about animating a photo

A first frame and, if you like, a last frame: PNG, JPG or WebP files up to 20 MB each. Only upload images you own or have permission to use.

Any whole length from 4 to 30 seconds, rendered at 480p, 768p, 1080p or 2K with native audio. 2K clips go up to 15 seconds.

Image to video keeps the aspect ratio of your photo. To pick a ratio yourself, generate from text instead.

The same as text to video: 4 credits per second at 480p, 6 at 768p, 8 at 1080p and 10 at 2K, so a 5-second 2K clip costs 50 credits. Clips longer than 15 seconds cost double per second.

Try MiniMax H3 image to video with your photo