How do you maintain consistent lighting and character position across multiple AI-generated shots?

Consistent lighting and character position come from reference anchoring, not prompt wording. Three methods that work:

1. Chain each shot off the previous one, starting from a wide establishing shot

2. Storyboard key frames as one vertical composite and animate from it

3. Lock character and location reference sheets before generating anything

Chain each shot off the previous one. Open every sequence with a wide establishing shot in your film's aspect ratio, then generate each subsequent shot with the previous clip attached as a spatial layout reference using Seedance 2.0 reference-to-video (R2V). The model reads the space, the light direction, and where each character stands from the reference and carries them forward, so shot 5 still matches shot 1. invideo is an agentic video creation tool with all the current models available, and the invideo agent holds that spatial layout across the whole chain — one documented production used exactly this to ship a 10-episode microdrama series in 3 days with a 3-person crew at $1000 per episode. Model choice matters here: Veo and Kling handle plenty of shot types, but for spatial and lighting continuity, route chained shots to Seedance 2.0's reference-to-video — the invideo agent does that routing for you.

Storyboard first for high-stakes scenes. Lay out the scene's key frames as a single vertical composite image — GPT-Image-2 generates these storyboard frames well — and have the invideo agent attach that composite to Seedance 2.0 and animate from it. Because lighting and blocking are fixed in the frames before any video generation, every shot inherits the same setup, you spend fewer credits, and your creative director can approve every frame before a single video credit is spent. The trade-off is real: you sacrifice the unexpected shots the model would otherwise generate freely, so reserve this for scenes where continuity outranks discovery.

Lock reference sheets before you generate, and iterate only from locked bases. Generate character sheets — front, side, and three-quarter angles plus facial close-ups — and location reference sheets for everything in parallel, before video production starts, then lock the finals as shared context assets. Every iteration after that — a costume change, a lighting shift, a new angle — starts from a locked base image, never from scratch, which is what stops lighting and positioning from drifting between generations. Locking after you've started generating frames is the common mistake; lock first.

One adjacent point worth knowing: because references live in central context, a mid-series change — a scar, a haircut — gets updated once and every subsequent shot inherits it automatically.

These are some of the ways to problem-solve this — what works depends on your shot.

Watch some of these to see what works for you:

Every sequence starts with a wide establishing shot. The Agent holds the spatial layout. Every shot after references the last - same space, same lighting, same character position.

— invideo's creative team