Yes — documented productions have expanded a 6×6 ft studio pool into open-ocean sequences and an 8×8 ft cave wall into a full cave exterior. The workflow: lock your practical shot as the anchor, upload stills from your own shoot as visual references, then generate wide environment shots that place the same action inside a larger world.
Start with your practical footage as the anchor and build the world outward from it. In one documented production, a filmmaker shot an emergence-from-water scene in a 6×6 ft studio pool — too small to hide the set edges, and the actor could only tread cold water for one to two minutes per take — then used AI to generate the red-ocean wides and shoreline shots that made the pool read as open sea. The same production built an 8×8 ft cave wall practically and generated the cave exterior around it. The principle holds in both cases: practical sets still need to be built — AI extends and augments physical production, it doesn't replace it.
Upload stills from your shoot as references before generating anything. Screenshot frames from your practical footage and feed them into the invideo agent — an agentic video creation tool with the current generation models available — so color, contrast, and look are anchored to your real material instead of a text prompt's guess. In the pool production, this reference-driven approach is what made the AI ocean match the studio water. Models like Seedance 2.0 that accept image references carry that anchoring directly into the generated environment shots.
Generate the wide environment shots, and let story continuity pick which ones. Ask for wides that place the same action in a larger world — the filmmaker chose a shoreline wide because the practical screenshot showed him on hands and knees coming out of the water, so a shoreline cut matched scene-to-scene. Choose extension shots by narrative continuity, not by which generation looks most impressive in isolation.
Keep the close-ups practical and the AI on the wides. Pull out rather than push in: AI-generated characters break down in close-up acting shots, so reserve AI for environment scale and keep faces and performance on your real footage. This division is exactly what makes a small-set extension invisible.
Iterate conversationally and reuse the established look. Work back and forth with the invideo agent — pin a generation you like and ask for another one with specific changes. Once the look and effect were established on the first cave shot, the second shot in the scene generated correctly on the very first attempt, because the session had internalized the visual language.
Blend the seams in post. Color grade the AI wides to match your practical footage, add sound design across the cut, and carry environmental continuity cues — matching haze between an exterior wide and your interior set shot, for example — so the transition between generated world and physical set reads as one location.
Watch some of these to see what works for you:
You definitely get so much more out of AI when you actually use your own footage as the references. You can make it look very close to something that looks real.
— Alex Arfaoui, filmmaker