Pixar-Style AI Story Videos with Seedance: Characters, Dialogue and an API Pipeline

What is a Pixar-style AI story video?

Pixar-style AI story videos are short episodes for TikTok, Reels and YouTube Shorts in which the same stylized 3D animated characters act out a small story with spoken lines. Four traits define the format:

• Vertical 9:16 episodes of 30 to 60 seconds, cut together from shots of 5 to 15 seconds.
• The same characters in every episode, in a stylized 3D animated look, so viewers recognize the hero from the first frame.
• Dialogue with lip-sync: one or two short lines per shot, spoken by a named character, in whatever language the channel uses.
• A simple arc: a hook, a setback, a turn and a payoff, often an underdog story, a family conflict or a side-business journey, sometimes ending on a product moment or a call to action.

Step 1: Build each character once

A character sheet is a single image that shows one character from the front, three-quarter and side, on a plain background, in the outfit the character always wears. Making one sheet per character before you write any shot is what keeps the hero identical across an episode: generate it as an image (Seedream on AiuniVid, in the browser or through the API) from a written description, and attach that exact image to every shot.

• Keep the design clearly stylized. Seedance refuses a recognizable real face as an input, so a photograph of a person, or a photoreal render of one, fails as a character reference. Illustrated and stylized characters are not affected. An image generated on AiuniVid is also accepted for 24 hours even when it looks realistic, but a sheet you will reuse for weeks should be stylized enough to pass on its own. On the website you can also save a character under Assets and use it from the API as an asset:// reference listed by GET /v1/characters.
• Describe the look instead of naming a studio. Words like premium 3D animated feature-film look, soft cinematic lighting, expressive faces and slightly exaggerated proportions are what the model acts on, and they keep the characters your own.

Character sheet of {HERO}, a {age} {build} young man, stylized 3D animated feature-film look.
Front view, three-quarter view and side view, full body, neutral pose, plain light-grey background.
Face: {round face, big brown eyes, thick eyebrows}. Hair: {short dark curls}.
Outfit: {black hoodie, loose black pants, white sneakers}, the same in every view.
Soft studio lighting, clean edges, no text, no other characters.

Step 2: Write each shot as a small script

Use reference-to-video and give every attached image exactly one job. In the prompt, refer to images by position (@image1, @image2) and say who each one is. Put the style line in every shot, then the action, then the lines of dialogue with the speaker named first.

@image1 is {HERO} only. @image2 is {MOM} only. Keep faces, hair, outfits and proportions identical to the references.
Premium stylized 3D animated feature-film look, soft cinematic lighting, expressive faces.
Vertical 9:16, {10} seconds, kitchen at night.
0–3s: {MOM} slams a bank statement on the table. Quick push-in on {HERO}.
3–10s: {HERO} looks up, calm.
{MOM}: "{You spent our savings on this?}"
{HERO}: "{Give me one month.}"
All dialogue in {English}. Only the named speaker moves their lips; everyone else keeps their mouth closed. No subtitles.

• One beat per shot. A shot that tries to hold a hook, an argument and a reveal usually lands none of them. Split it and cut the pieces together.
• Keep lines short and give each one a named speaker. Two speakers per shot is a practical maximum.
• Say the language when the dialogue is not in English, and write the line in that language.
• Inner thoughts are voice-over. Write that the character’s mouth stays closed, or the model will animate the lips.
• More fill-in templates are in Seedance prompt examples, and the full method for holding a face across a series is in consistent characters in Seedance.

Step 3: Chain shots into an episode

To chain AI video shots into one episode, use the last frame of each finished clip as the first frame of the next shot, and keep the character sheets attached to every shot. Every finished AiuniVid task returns that frame as last_frame_url. In image-to-video it becomes the opening frame; in reference-to-video you can attach it next to the character sheets and write that the clip starts exactly on that image. A 60-second episode takes four 15-second shots or six 10-second shots, so there are three to five cuts to chain, and the sheets are what stop faces from drifting across them.

How first and last frames work is on the image-to-video page; how references are numbered is on the reference-to-video page.

Step 4: Pick the model, and what an episode costs

Seedance 2.0 Mini is the cheapest Seedance tier for story reels: 4–15 seconds per shot, 480p or 720p, and up to 9 reference images, enough for several characters and a location. The per-second rate matters more here than for one-off clips, because story reels are made in volume. Move up to Seedance 2.0 for 1080p, or to Seedance 2.5 for single takes of up to 30 seconds.

One credit is $0.01. An episode here is 4 shots of 15 seconds; failed generations are not charged, and every job is quoted before it runs. Reference images do not change the price; a reference video is billed for its own seconds as well. Current rates for every model are on the pricing page.

If an episode needs 1080p, preview each shot on Seedance 2.5 as a 480p draft and render only the keepers at 1080p: how draft mode works.

Step 5: Run it through the API

A story-reel pipeline on the AiuniVid API generates character sheets with POST /v1/images/generations, creates every shot as an asynchronous task with POST /v1/videos/generations, and receives each finished shot at a callback_url. Four rules keep it cheap and reliable at volume:

• Use a webhook instead of tight polling. Pass callback_url and the result is pushed to you when the shot finishes; polling every second burns requests and rate limit for nothing.
• Send an Idempotency-Key with every create request and reuse it when you retry after a timeout, so a lost response never pays for the same shot twice.
• Quote before a batch. The quote endpoint returns the exact credits for a request without creating anything; check the batch total against your balance, because a create request with too few credits returns 402.
• Do not retry a 422 real_person_detected. The same input fails the same way; swap in a stylized sheet instead. Create requests are limited to 10 per minute per key; spread a batch rather than bursting it.

curl --request POST \
  --url https://aiunivid.com/api/open/v1/videos/generations \
  --header "Authorization: Bearer $AIUNIVID_API_KEY" \
  --header "Idempotency-Key: $(uuidgen)" \
  --header "Content-Type: application/json" \
  --data '{
    "model": "byteplus-seedance-2.0-mini-reference-to-video",
    "prompt": "@image1 is HERO only. @image2 is MOM only. ...",
    "image_urls": ["https://example.com/hero-sheet.png", "https://example.com/mom-sheet.png"],
    "duration": 10,
    "quality": "480p",
    "aspect_ratio": "9:16",
    "callback_url": "https://your-app.example.com/webhooks/aiunivid"
  }'

Every field, limit and error is in the API docs. You can also drive the same steps from Claude, ChatGPT or Cursor through the AiuniVid MCP server.

When something goes wrong

Most failed 3D animated story shots come from one of five causes, and all five are fixed in the reference or the prompt rather than by retrying the same request.

Frequently asked questions

Pixar and Disney are trademarks of their respective owners. AiuniVid is not affiliated with or endorsed by them; Pixar-style is used here only to describe a visual look.