Using AI to generate animated shorts: storyboard to assembled video

Generating "an AI video" is easy. Generating an animated short with the same character across every shot, scene continuity, and a story that lands is a different beast.

This is the breakdown of how Lumora's video pipeline produces a finished animated short — six stages, the same character bible as novels and comics, and clear cost gates at each step.

The six-stage pipeline

A video project in Lumora moves through:

1. Preparation — characters, premise, target length.
2. Planning — chapter outline (same shape as novels/comics).
3. Storyboard — shot-by-shot description: framing, motion, mood.
4. Keyframes — the start frame of each shot, generated as a still image.
5. Animation — each shot animated from its keyframe.
6. Assembly — clips stitched into a single MP4 with timing and (optional) audio.

Each stage is gated. You approve or regenerate before paying for the next one.

Stage 1: Same character bible, new pressure

Character consistency is brutal in video. A novel can describe a character's face once and you forget the details by chapter 3. A comic can keep them recognizable across 24 stills. A 90-second video can have 30+ shots, each one a fresh chance for the model to drift.

Lumora's mitigation: the same character bible you used for a novel or comic feeds the video pipeline directly. The reference description ("Iria, 31, copper hair tied back, beekeeper's apron…") plus any uploaded reference images get pinned to every keyframe and every shot.

If you're starting fresh, take the time to write a 3–4 line character description per main character. It will pay back across every shot.

Stage 2: Plan it like a short film, not a music video

The planning step gives you a chapter outline — same as the other formats. For video, treat each chapter as a scene: roughly 10–20 seconds of finished video, 3–6 shots.

Common mistakes:

• Too many scenes for a 60-second short. 4–6 scenes is plenty. More turns into a montage.
• No establishing shot. The audience needs 1.5 seconds of "where are we" before action.
• All medium close-ups. Mix wide / medium / close. The pipeline supports it; use it.

Stage 3: The storyboard is the highest-leverage edit

The storyboard step generates one card per shot:

• Shot number, duration, framing (wide/medium/close)
• Motion (static, slow push, pan left, etc.)
• Mood / lighting cue
• Subject and action

This is the cheapest step and the one that determines everything downstream. Read every shot card. If the framing is wrong, fix it now — fixing it after keyframes are generated costs you the keyframes.