Cartoon and stylised output is the easiest thing to ask of an AI video model, for a structural reason: the uncanny valley is a photorealism problem. Flat shading, simplified anatomy and exaggerated motion are a visual grammar that absorbs exactly the artefacts that ruin a realistic clip. Which makes this the friendliest entry point in the whole category.
The hard part is not the look. It is keeping the same character across shots.
TL;DR
- Free tiers: watermark, roughly 5 seconds, low resolution, slow queue
- Cheapest paid route: Wan 3.0 at 480p, 24 credits per 5 seconds
- Stylised beats realistic because flat shading hides the failure modes
- Character consistency is the real problem, and the fix is to lock the design in a still first
- Reference-to-video on Wan 3.0 Prime at 34 credits is the cheapest way to carry a design forward
Why cartoon is the easy mode
Flat colour hides inconsistency. Photorealistic lighting must stay physically coherent frame to frame. Two or three flat tones do not.
Simplified hands. Hands are the reliable AI failure, and cartoon hands are already simplified shapes with four fingers and no detailed knuckles.
Exaggerated motion is the style. Squash, stretch and impossible arcs read as craft rather than as error.
Held frames are legitimate. Limited animation is a real technique, so a clip that does not move constantly looks intentional.
The workflow that works
Do not prompt straight to video. Character consistency is the hardest problem in the medium and prompting to video gives the model freedom you do not want.
- Design the character as a still. Seedream 5 at 6 credits. Iterate until the design is right; twenty attempts cost 120 credits, about $1.20.
- Write the character block. A paragraph describing the design exactly: proportions, palette, clothing, line weight. Paste it unchanged into every subsequent prompt.
- Animate from that still. Wan 3.0 at 480p, 24 credits per 5 seconds, using the approved frame as the first frame.
- Reuse the same frame every time. Each new shot starts from the approved design, not from the previous clip.
About 30 credits per shot, thirty cents.
What it costs
Canon 8frame prices, a credit is $0.01 at pack rate:
| Route | Clip or image | Credits |
|---|---|---|
| Character design still | Seedream 5 | 6 |
| Illustrative, non-photoreal still | Reve | 34 |
| Animate, 5s at 480p | Wan 3.0 | 24 |
| Animate at any resolution, 5s | Wan 3.0 Prime (reference-to-video) | 34 |
| Animate, 5s at 720p | Wan 2.2 | 50 |
| With generated audio, 8s | Veo 3.1 Lite | 54 |
| Controlled camera move, 5s | Kling v3 Standard silent | 57 |
| Long single take, 30s | Wan 3.0 Prime | 203 |
| Assemble the shots | Merge Videos | 10 |
A six-shot cartoon sequence with a designed character: roughly 300 credits, about $3.
Where it still breaks
The character drifts. Even with a reference frame, side and three-quarter views of the same design will differ. Build sequences from shots that reuse similar angles.
Two characters interacting. The reliable failure. Bodies pass through each other and neither design holds.
Lip sync to dialogue. Not solved in this workflow. Either avoid visible speech, cut away during lines, or keep mouths simple and rhythmic rather than phonetic.
Text in frame. Signs and titles garble once they move. Add type in the edit.
Long-form. A full animated short needs consistency, dialogue and staging together, and those are unsolved as a set. Short sequences are the realistic scope.
The copyright line
Do not reproduce an existing studio's characters. They are copyrighted and immediately recognisable, and generating them for anything public is a straightforward infringement question rather than an interesting edge case. The same applies to prompting a named living artist's style.
Describing a technique is legitimate. Naming a property or a person to copy is not.
FAQ
Can I turn a photo of myself into a cartoon video? Technically yes. For a real person, only with their permission, and expect the likeness to drift.
Why does my character keep changing? Because nothing carries the design forward unless you make it. Lock it in a still and reuse that frame as the first frame of every shot.
Is 480p enough for cartoon? Flat-shaded output survives low resolution better than photorealistic footage, so usually yes. Generate at delivery resolution, since there is no video upscaling here.
What is the cheapest full workflow? Design at 6 credits, animate at 24, merge at 10.
Design the character once, then reuse that frame everywhere. The 8frame canvas is free and unlimited, and generation is paid from $19/month.