Can AI animate manga panels? Yes, one panel per clip

Yes, AI can animate manga panels, one panel at a time. Each finished panel becomes the first frame of a short image-to-video clip, and the clips are joined in reading order into a motion comic with transitions and sound. The panel is the clip's opening image; what moves after that comes from your prompt and the model.

The fields and limits come from Sume's Video generation and Timeline 1.0 docs and the model catalog behind GET /v1/videos/models, read on 2026-09-27. Anything described as current behavior is read from Sume's API code.

How do I animate one manga panel?

Send the panel as the first_frame in frame_images on POST /v1/videos, and describe the motion in the prompt: what moves, how the camera moves, and the light. The panel must sit at a public HTTPS URL. A clip runs 2–30 seconds depending on the model, and GET /v1/videos/models lists each model's supported_durations.

{
  "model": "wan-3.0",
  "prompt": "Her hair and scarf lift in a strong wind, speed lines streak past, slow push-in on her face",
  "frame_images": [
    {
      "type": "image_url",
      "image_url": { "url": "https://example.com/manga/ch3-panel-07.png" },
      "frame_type": "first_frame"
    }
  ],
  "aspect_ratio": "3:4",
  "duration": 4
}

Which panel shapes can an AI video model animate?

The shapes a model lists in supported_aspect_ratios, since aspect_ratio must be a value from that list. No model lists 2:3 or 4:5, so a tall panel needs 3:4 or 9:16, and a wide one 16:9 or 21:9. The docs don't say how a model treats a first frame whose shape differs from aspect_ratio, so crop or pad each panel to a listed ratio before you send it, and set aspect_ratio to match.

Can I keep a character consistent across panels?

Each panel carries its own drawing of the character into its clip, because the clip opens on it. What you cannot do is pin a panel and add character reference images in the same request: when a request has both frame_images and input_references, frame_images takes precedence and the request is treated as image-to-video, so the references are dropped. The docs promise no identical look from clip to clip; Consistent character across AI video shots compares the inputs.

How do I turn the panels into a motion comic?

Timeline 1.0 joins the clips in reading order in one render: 1–200 slots, each with a start and a duration, and a fade, wipe, slide, or dissolve of up to 1 second between slots. For clips whose shape differs from the output, fit takes cover (the default), contain, stretch, or blur.

Every slot must already be this workspace's media.sume.com file. Generated clips are; your own panel files are not, and the public API documents no upload route, so each panel reaches the timeline as its clip. The sound comes from the audio spine: narration from text to speech, a track from Sume's Music Router, or audio.mode: "silence" for none. In current code a clip's own audio is not mixed in. Output edges run 256–2160 pixels, 1080×1920 by default. How to assemble a long-form video covers the request.

What are the limits?

• Lettering and speech balloons: nothing in the docs promises they stay legible once the panel moves.
• Art style: the clip opens on your panel, and the docs make no promise about how closely the motion after it keeps the style.
• Aspect ratios: only the values each model lists, with no 2:3 or 4:5.
• Clip length: 2–30 seconds per generation, depending on the model.
• Frame URLs must be public HTTPS; signed URLs are rejected.