English Learning Animation

Create a coherent, original short-form English lesson. Prioritize a watchable scene over a slide deck with narration.

Workflow

1. Define one communicative outcome and write an English-only script with 3–5 usable phrases, a natural dialogue, and a brief repeat-after-me close. Do not choose a target runtime first. Let the generated speech, necessary pauses, cover, and recap determine the final duration. Many lessons will naturally land near 25–45 seconds, but this is not a quota.
2. Build a shot list before generating visuals. Add a semantic_contract to script.json: topic, setting, visual brief, required scene tags, and stale terms that must never appear. Every scene needs semantic_tags. Give every voice segment a stable semantic owner such as customer, barista, or narrator; do not use gender as the long-term character identity.
3. Use Qwen3-TTS VoiceDesign when no reference audio exists. Create a short audition for every recurring role first; do not reuse one voice for multiple characters. Lock each approved role's voice_profile and add line-specific performance direction.
4. Generate an empty background plate and separate transparent character/prop cutouts that visibly match the current setting. A hotel lobby cannot stand in for a subway station, restaurant, or attraction. Use a layered animation system such as paper-collage-remotion; never animate a single flattened illustration as the whole video.
5. Place audio using actual generated durations, then make visual changes at segment starts, phrase beats, and turn changes. Never add dead air or extend scenes merely to reach a round-number runtime. During dialogue, keep the speaker visually primary; during narration, use an English phrase card or semantic graphic rather than pretending a character is speaking.
6. Render only after passing the quality gates below.

Production Starter

Initialize a new project from the approved layered-animation baseline: