Can AI animate old photos? How to bring a print to life

Yes. AI can animate an old photo: scan the print, put the scan at a public link, and send it to an image-to-video model as the clip's first frame, with a prompt that describes the movement. The clip opens on your photo, and the model generates the seconds after it.

The steps below use Sume's video API. Facts come from the Video generation and Media inputs docs, the OpenAPI document behind the Sume API reference, and Sume's video catalog, read on 2026-09-27.

How do I animate an old photo, step by step?

• Scan the print, or photograph it flat, and crop the scan to the aspect ratio you will ask for (next section), so the first frame and the clip have the same shape.
• Put the scan at a public HTTPS URL. Signed, private, localhost, and non-HTTPS links are refused.
• If the scan is small, you can enlarge it first with Image Upscale 1.0 (POST /v1/image-upscale-1.0/upscale), listed at $0.20 per image on API pricing. It takes the scan's public HTTPS URL, and its result is a Sume-hosted artifact on media.sume.com that you can send as the first frame; the AI image upscaler API covers the request.
• Send the image to POST /v1/videos in frame_images as the first_frame, with a prompt for the movement you want, such as a blink, a slight smile, or a slow turn of the head.
• Poll the job until it completes, download the clip with your API key, and watch it before you share it.

curl -X POST https://api.sume.com/v1/videos \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: old-photo-001" \
  -d '{
    "model": "seedance-2",
    "prompt": "The couple in the old photograph blink and smile slightly",
    "frame_images": [
      {
        "type": "image_url",
        "image_url": { "url": "https://example.com/old-photo-scan.jpg" },
        "frame_type": "first_frame"
      }
    ],
    "aspect_ratio": "3:4",
    "duration": 5
  }'

Which models fit the shape of my print?

Each video model offers a fixed set of aspect ratios, and only some take 4:3 and 3:4. No model lists 3:2 or 2:3, so crop a print of that shape to the nearest ratio the model lists. In current code, an aspect_ratio the model doesn't list is refused with a 400.

Will the people in the photo still look the same?

The clip opens on your scan as its first frame. Every frame after that is generated, and no setting is documented that keeps a face identical through the clip, so watch the whole clip before you share it, and generate another take if you don't like it.

What can't it do?

• Restore or colorize the photo. The video models are not documented to do either; the clip starts from the scan as you send it.
• Make someone speak. Sume's docs say video models do not lip-sync to generated speech or a later voice-over; for a talking photo, use lip sync with Sume-hosted audio such as a text-to-speech clip, covered in the lip sync API guide.
• Run long. One clip is 2 to 30 seconds depending on the model; How to make a picture move with AI compares the other ways to animate a still.