Animate a pencil sketch or storyboard panel as the first frame

To animate a pencil sketch or a storyboard panel, send it as the first_frame in frame_images on POST /v1/videos, and describe the motion in the prompt. If the next panel is ready, send it as the last_frame on a model that takes an end frame, so the clip travels from one panel to the next. A storyboard becomes an animatic this way, one clip per pair of panels.

A rough drawing is a good input for a reason: it carries composition and not detail. The model has freedom to fill the lines, so decide before you start whether you want the final clip to look like a drawing that moves, or like a finished scene that grew from the drawing. The first needs the style words in every prompt. The second needs a prompt that describes the finished look, and you should expect the lines to vanish.

The prompt should say what moves and what keeps its drawn look. Without that, a model tends to read a sketch as a rough photograph and fill in realistic texture.

Which models take an end frame

The end frame is what turns a single panel into a transition. With only a first frame, the model decides where the shot goes. With both, you decide the start and the finish and the model supplies the movement between them, which suits a storyboard where the next panel is already drawn. Keep the two panels close in composition, with the same character size and the same horizon, or the model has to invent a large change. The catalog lists first and last frame image-to-video on these models. A model that lists only a first frame can still animate a single panel.

Writing the prompt for a drawing

Treat the prompt as a short direction to an animator who has only the drawing. Name the subject once, the action once and the camera once. Long prompts fight with a drawing, which already says a lot. Describe the motion and the camera, and name the style that has to stay. The style words belong in the prompt because the model sees only a picture and can drift away from a line drawing.

• Say the style: "pencil sketch, graphite lines, paper texture stays visible".
• Say one action per clip: "the cyclist leans into the turn".
• Say the camera: static, slow push-in or a pan, never several moves.
• Leave out quality words that pull toward photography.

Request from panel one to panel two

This body travels from the first panel to the second. The aspect ratio of both panels should match the aspect_ratio you request, so crop them to the same ratio before you upload.

curl -X POST "https://api.sume.com/v1/videos" \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "wan-3.0",
    "prompt": "Pencil sketch stays visible. The cyclist leans into the turn.",
    "duration": 4,
    "resolution": "720p",
    "frame_images": [
      {"type": "image_url", "frame_type": "first_frame",
       "image_url": {"url": "https://example.com/panel-01.png"}},
      {"type": "image_url", "frame_type": "last_frame",
       "image_url": {"url": "https://example.com/panel-02.png"}}
    ]
  }'

Planning a whole board

Scan the panels before you upload. Remove notes, arrows and frame numbers from the image, because a model may animate them as if they were part of the scene. Keep the file at a reasonable size and on a public HTTPS address. A phone photo of a paper sketch usually needs a crop to remove the table at the edges and a lift of the contrast so the lines are clear.

Set the length of each clip from the beat of the story, not from the panel count. A quiet establishing panel can run 5 or 6 seconds, and a fast action panel can run 2 or 3 on Wan 3.0. Keep the model the same for the whole board so that the line quality matches, and test one pair of panels before running the rest. When the first and last frame disagree strongly, the model has to invent the bridge, so keep neighboring panels close in composition. For the base rule on pinning a frame, see the clip that does not start on your photo.