How to make AI skits with a recurring cast
Keep the world recognizable. Change the situation.
A repeatable AI skit workflow separates the story from the reusable scene: write one clear situation, keep the characters and room consistent, approve the voices, then render and review the episode. Formato Skits uses a code-rendered low-poly scene and generated voices. It does not need a new video-generation prompt for every shot.
№003Write the exchange before building the scene
The format has two roles: an owner with a painfully specific problem, and a helper whose matter-of-fact response makes the problem clearer. Start with a situation a customer can recognize. Keep the product claim factual; a comic exaggeration should not become a false promise about what the brand does.
Write a short deadpan skit for [brand]. Owner: [person and problem]. Helper: [character and voice]. The owner confesses one specific disaster. The helper responds with flat facts, then offers one believable save. Use a single night room. Give each line a purpose, mark pauses, and stop for script approval. Do not invent product capabilities.
What stays the same across episodes
Scroll sideways to see all columns.
| Keep stable | Change deliberately |
|---|---|
| Owner and helper rigs | Facial expressions, poses and reactions |
| Room layout and visual style | Props, shelf stock and a small amount of set dressing |
| Character roles | The particular problem and punchline |
| Voice identity | The performance and selected take of each line |
A practical episode sequence
Approve the script
Read the dialogue aloud. Remove explanations the scene already communicates. Make sure the setup and payoff still work without the brand name.
Check the cast and room
Use the still lab to inspect faces, costume and framing before the full render. This is a cheap place to catch a character mismatch.
Pick performances
Generate voice takes, listen in context and choose them. Do not judge timing from isolated lines alone.
Render a review copy
Check the first few seconds, pauses, captions and phone-speaker sound. Fix the relevant stage instead of rerunning every paid call.
What Skits adds
Skits v0.2.0 includes six owners, a night room, a helper character builder, a still lab and the production/review process. ElevenLabs is required for the voice stage; Gemini and fal are optional. The example is in English. Review any new language’s performance and caption handling before committing to a series.
Inspect the real film, included parts and requirements on the Skits page.
Skits · $49 →Questions
Does it generate every frame with an AI video model?
The core Skits scene is rendered from code with three.js. Voices use ElevenLabs; optional tools support other parts of the workflow.
Can I use my own brand character?
The pack accepts a brand helper module or a character-builder specification. Inspect and approve the resulting character before making a series.
Keep reading
- A video brief your coding agent can useOne clear job for the film, before the first render.Read →
- What does an AI video workflow actually cost?The download price is only one part of the budget.Read →
- Fix the failed stage, not the whole filmA troubleshooting checklist for local agent video workflows.Read →