How do you create location reference sheets for an AI video series?

Create location reference sheets by generating multiple views of every location — a wide establishing view, mid views, and key detail views — in one parallel pass alongside your character sheets, before any video generation. Lock the finals, store them as central context so every episode agent inherits them, and update mid-series changes once in that single source.

Build all your locations in one pass, before you generate a single video frame — the same pre-production discipline you apply to characters. invideo is an agentic video creation tool, and the workflow below runs through the invideo agent, which holds your reference sheets as project context and applies them across episodes.

1. Generate every location in parallel with your cast. Run one image-generation pass that covers all characters and all locations together, then stop and lock before production starts. "All characters, all angles - front, side, three-quarter - generated in parallel. Same for every location. Lock the finals," is how invideo's creative team describes the pass. Image models like GPT-Image-2 or Nano Banana handle the still generation inside invideo, so the reference build and the video production live in one project.

2. Give each location multiple views, not one hero image. Mirror the multi-angle logic of character sheets: a wide view that establishes the full spatial layout, mid views of the areas where scenes actually play, and detail views of props or surfaces the camera will land on. The wide view does double duty later as the spatial anchor for generation.

3. Lock the finals and review them before spending video credits. Approve each location sheet as final, then treat those images as immutable bases — every later iteration (a lighting shift, a new angle, a redress) starts from the locked base image, never from scratch. Locking before generation, not after, is the step that prevents environment drift from creeping in episode by episode.

4. Store the sheets as central, shared context. Upload the locked location sheets into your project's central context — in invideo, the notebook that your showrunner agent holds — so every per-episode sub-agent and every team member pulls from the same source. One documented production ran a 10-episode series this way with a 3-person crew in 3 days at roughly $1,000 per episode, with each collaborator's agent instance loaded with the same locked show bible, scripts, character sheets, and location references.

5. Anchor each sequence to the location's wide view. Start every scene with a wide establishing shot built from the location sheet, then generate each subsequent shot with the previous one as a spatial layout reference using Seedance 2.0 reference-to-video. The invideo agent holds that layout shot to shot, so space, lighting, and character position stay consistent through the sequence.

6. Update mid-series changes once, centrally. If a location changes partway through the series — a set redress, new signage, seasonal shift — update the sheet once in the central context and every subsequent episode inherits it automatically, the same way a character's scar or haircut propagates.

7. Skip sheets for throwaway environments. For insert shots and passing montage locations, don't generate reference images at all — prompting the model directly from the script description is sufficient. Reserve the sheet-building pass for locations that recur across scenes or episodes.

Watch some of these to see what works for you:

All characters, all angles - front, side, three-quarter - generated in parallel. Same for every location. Lock the finals.

— invideo's creative team