How do you create a multi-angle character turnaround sheet for consistent AI video generation?

A character turnaround sheet for AI video needs four angles — front, three-quarter, side, and back — plus a face close-up and a mid-angle close-up, generated at 4K with the character's hands empty. Cast the face first, generate the sheet in Nano Banana Pro, produce four variations, lock the best one, and attach it to every video generation prompt.

Decide your angle set before touching a model: front, three-quarter, side, and back, plus a face close-up and a mid-angle close-up — the close-up panels are what keep small details like scars and accessories consistent, because video models reproduce only what the sheet actually shows. invideo is an agentic video creation tool with all the current image and video models — Recraft, Nano Banana, GPT-Image-2 for images; Seedance 2.0, Veo, Kling for video — so the full workflow below runs in one place.

1. Brief the invideo agent on the character first. Give it the character description, era, costume, and any defining props before generating a single image — in one documented session the invideo agent itself asked for exactly these answers before building assets, because they change every frame downstream.

2. Cast the face before the turnaround. Run the same character prompt on two image models in parallel — for example Recraft and GPT-Image-2 — and pick the aesthetic you prefer; one production ran its casting through a dedicated casting agent testing two models simultaneously to speed iteration. Recraft is strong for photoreal faces because it renders skin imperfections — pores, lines, stubble — and generate the chosen portrait at 4K so the sheet has a high-detail identity anchor.

3. Generate the turnaround sheet in Nano Banana Pro at 4K. Feed it the locked portrait and request the four angles plus the face and mid-angle close-ups in one sheet; Nano Banana Pro outperforms Nano Banana 2 for character sheet fidelity. Remove any objects from the character's hands before generating — held props drift between angles, so generate each prop as its own separate reference instead (one production covered four characters and one prop with 11 reference images total). If you already have a 3D model of the character, rendering front, side, profile, and back views works as the sheet too — one 70-second short used exactly that.

4. Generate four variations and lock one. Produce four versions of the sheet, select the best, and formally lock it before any video generation begins — locking references upfront is the step that prevents consistency problems across the rest of the film. Budget roughly 5 generation attempts to lock one character, about $9.78 per character at documented rates.

5. Attach the locked sheet to every video generation. The invideo agent stores the sheet in project context and attaches it to each shot prompt; on the video side, Seedance 2.0 reference-to-video accepts character references alongside location references, so identity carries across clips rather than resetting every generation. This is the mechanism that held two characters visually identical across every scene of a 70-second short film with no LoRA fine-tuning. If a continuity error shows up in a shot later, don't re-roll the shot — ask the invideo agent to trace the error to the specific sheet panel, fix it there, and the corrected sheet propagates to all subsequent generations.

Two edge cases worth planning for. If the character's appearance evolves across the story — costume changes, accumulating items — build a distinct sheet per beat: one production's character picked up a new trinket in every city, so the team made a separate character sheet for each sequence. And if the sheet involves two characters in physical contact (one carrying another, shared props), hand-sketch the exact configuration and upload the drawing as a reference; the invideo agent feeds it to the image model to produce an accurate fused sheet where text prompting alone won't.

Watch some of these to see what works for you:

the AI always needs to see what the character is exactly, right? Or else it'll kind of hallucinate and imagine something that's under the cap. So, we don't want to do that. We always want the character to be seen as we see it on the character sheet.

— invideo's creative team