1. Write a production brief
Define the audience, platform, target duration, aspect ratio, story promise, and one desired action. Add non-negotiable facts and prohibited claims. A tight brief keeps later model and style choices from drifting.
2. Shape the script for pictures
Convert narration into beats that can be shown. Mark each location, character, action, emotional turn, voice line, and transition. If a paragraph cannot be visualized clearly, rewrite it before spending credits.
Official how-to guide ↗3. Build and approve characters
Create a compact identity sheet for each recurring character: face, hair, age range, wardrobe, accessories, palette, posture, and emotional baseline. Generate neutral views first and reject inconsistent details before the character enters several scenes.
4. Turn beats into a storyboard
For every scene, specify subject, action, setting, camera, duration, light, sound, and the exact continuity anchor from the previous scene. Use the storyboard to expose missing transitions and expensive scenes before rendering.
5. Choose a model by shot
Do not select one model for an entire project by habit. Compare Seedance, Kling, Veo, and other currently available options on the hardest representative shot. Verify live model access because the platform catalog and plans can change.
6. Review and regenerate narrowly
Check identity, hands, props, spatial logic, camera intent, lip sync, text, and usable duration. When a scene fails, change one variable at a time. Rebuilding the entire episode makes it difficult to learn which instruction helped.
7. Finish sound, captions, and delivery
Listen for names, timing, clipping, and music conflicts. Review captions on a phone, verify rights to every upload and voice, then export a platform-specific master. Keep settings and approvals with the final file for later episodes.