The robust workflow is: photograph or export the drawing cleanly, lock its visual invariants in the prompt, request one motion verb, and test the result against frame-level assertions. This is image-to-video generation, not skeletal rigging.
Input contract
Prepare a PNG or JPEG of artwork you own or have permission to animate.
required:
- sharp silhouette
- even lighting
- complete subject inside the frame
locked:
- line weight
- palette
- paper texture
- character proportions
variable:
- one subject action
- one camera rule
Open the Draw to Video workspace. Today it selects Seedance 1.5 Pro and exposes a prompt, source-image upload, optional last frame, duration, aspect ratio, and Generate. The checked default is five seconds at 16:9 with a paid estimate of 7 credits.
Prompt fixture
Preserve the exact black ink line weight, flat red-and-cream palette,
paper grain, face design, and proportions from the source drawing.
The fox blinks once and lifts its tail gently.
Locked camera. No zoom. No new shadows, text, objects, or 3D shading.
End close to the starting pose so the clip can loop.
Treat this like a fixture. Change the subject and action, but keep the invariant section explicit.
Execution
- Crop the drawing to the output ratio before upload.
- Choose a motion preset close to the intended action.
- Paste the style-lock prompt.
- Keep the first run at five seconds.
- Verify the displayed credits and Generate.
Assertions
assert line_weight_variation < noticeable
assert palette_has_no_new_colors
assert character_identity_is_stable
assert no_new_objects
assert final_frame_supports_clean_cut_or_loop
Review at 25%, 50%, and 100% playback speed. Fast playback hides shimmer; paused frames expose it.
Troubleshooting matrix
| Symptom | Likely cause | Smallest fix |
|---|---|---|
| glossy 3D look | material language is vague | name “flat ink on paper”; exclude 3D shading |
| vibrating edges | source is soft or motion is large | sharpen input; reduce action |
| changing face | too many simultaneous changes | lock face; request one action |
| loop pops | final pose is unrelated | ask to return near the starting pose |
Limits
The model synthesizes intermediate frames and can repaint the artist’s marks. It does not provide exact keyframes, layers, or deterministic reruns. Do not animate another artist’s work without permission, and disclose AI-generated motion when publishing.
When the invariants and assertions are written down, run the exact drawing animator and confirm the current model, duration, ratio, and credit estimate first.