Can AI agents auto-generate character sheets and location sheets from a reference ad or video?

Yes — upload a reference ad to the invideo agent and it builds the character sheet, the location sheet, and a translated voiceover just from watching the video, with no manual brief. It detects every cut, extracts one frame per scene, and turns those frames into structured reference assets you lock before any new shot is generated.

Here is how the auto-extraction actually runs inside the invideo agent, step by step.

1. Upload the reference ad and let the agent decode it. Drop the winning ad (or any reference video) into the chat. The invideo agent watches the full clip, transcribes the voiceover, and detects every cut — in one documented localization, it found 9 cuts in the reference automatically and pulled one representative frame per scene. It also segments scene changes on its own (in one case identifying that the model wore one outfit in the first half, walked off-frame, and returned in a second outfit). This is the source material for both sheets.

2. The agent generates the character sheet from those frames. From the extracted frames, a storyboard sub-agent or character sub-agent generates a multi-angle character sheet — front, side, back — for each person it identifies, using GPT-Image-2 (the strongest current model for reference-attached character generation). Each sheet takes roughly 5 minutes to generate. You review, iterate granularly until the face holds, then lock the version by referencing its exact version number so every later generation pulls from that locked sheet. For batch localization across markets, name the character images after the target language (Spanish.png, French.png) and the agent auto-assigns each to the correct market ad.

3. The agent generates the location sheet the same way. From the same extracted frames, the agent rebuilds each location as a keyframe — the opening frame of the shot, framed exactly how the scene starts. GPT-Image-2 is again the preferred model for environment/location sheets because it accepts reference attachments and renders realistic interiors and exteriors. Review the before-and-after location states side-by-side in a grid to confirm lighting and color palette stay consistent before locking. Hridaye, invideo's creative director, frames the principle plainly: "The agent builds the character sheet, location sheet, and translates the video just by reading my reference ad."

4. The agent returns a structured change plan, not just visuals. Before generating anything, ask the invideo agent "what's the best way to do this?" — it returns an estimated generation breakdown: how many reference sheets, how many video clips, how many UI screens, and roughly what you'll spend. Across documented localization runs, that plan typically resolves to ~4 reference sheets per market, ~13 video-clip regenerations per ad, and ~5 app UI screens regenerated — all derived from the reference ad's structure, with the 7-shot beat order from the original preserved.

5. One caveat: strip burned-in captions before generation. If the reference ad has on-screen captions baked in, don't re-attach the video file when generating new shots — the underlying model will reproduce the old captions in your new frames. Instead, drive generation only from the pre-built character sheet, location sheet, and shot-flow breakdown the agent has already extracted.

Beyond the auto-extraction itself: full sheets aren't always necessary. For a simple single-location, single-character UGC ad, skipping the formal character-sheet and location-sheet step and going straight to keyframe iteration is faster. The sheets pay off when you're scaling one winning ad across products, markets, or languages — the documented localization workflow produced 6 fully localized UGC ads (3 ads × 2 markets) for $425 total ($70/ad, ~2.5 hours per ad) using exactly this auto-generated-sheet pipeline.

Watch some of these to see what works for you:

The agent builds the character sheet, location sheet, and translates the video just by reading my reference ad.

— Hridaye, invideo's creative director