How do you decide which shots to generate with AI in a hybrid live-action film?

Four rules decide which shots go to AI in a hybrid live-action film:

1. Story continuity — generate the shots that bridge your practical footage

2. Performance distance — wide environments to AI, close-up acting stays on camera

3. Build outward from a locked practical shot

4. Reserve rough AI output for fast inserts

Decide by story continuity first. Pick AI shots for narrative continuity, not visual impressiveness — the shots worth generating are the ones your edit needs to connect scene A to scene B. In one hybrid production, the filmmaker chose a shoreline wide as the AI shot because his practical footage showed him on hands and knees coming out of the water; the generated shot earned its place because the two cuts felt like one continuous scene. List the gaps in your edit before you generate anything — that gap list is your AI shot list. Carry environmental continuity cues across the seam too, like matching haze between an exterior AI shot and a practical interior, or the transition reads as two different films.

Assign by performance distance. Wide environment shots are AI candidates; any shot where a character does visible acting stays practical. Pull out rather than push in — the wider the frame, the less the model has to fake about human performance, and the less a viewer can catch.

Build outward from locked practical shots. When your practical setup is physically confined, lock that footage as the anchor and generate the wider world around it. One production shot an emergence-from-water scene in a 6x6 ft studio pool — too small to hide the set edges, with an actor who could tread cold water for only one to two minutes — so the team generated the open ocean, shoreline, and cave-exterior wides around the locked practical takes. Inside the invideo agent, upload screenshots from your own shoot as references before generating, so color, contrast, and look stay anchored to your practical material; the invideo agent routes those environment generations to models like Seedance 2.0, which produced the ocean and shoreline wides in that production. Once the first AI shot in a scene establishes the look and tone, subsequent shots in the same scene come faster and more accurately — in that production the second cave shot landed on its first generation.

Keep practical whatever the actor physically touches. AI extends physical production rather than replacing it: the same team still built an 8x8 ft cave wall set, then used AI to layer visual effects on top of it. Shots involving physical contact, texture, or close interaction get built and shot for real.

Reserve rough AI output for fast inserts. AI footage that looks artificial at normal speed can still work as short cutaways — speed ramp it 1,000% to 5,000% in Premiere Pro and splice the fragments into a quick, energetic sequence. Expect to pull the best seconds from several imperfect generations rather than one perfect clip, then bind AI and practical footage with a matching color grade and sound design — sound is the multiplier that makes AI inserts feel physical next to real footage.

These rules cover most shot decisions in a hybrid workflow — the final test is always the edit: if the AI shot serves story continuity, it stays.

Watch some of these to see what works for you:

If you push into this shot, or if the shot is too close, and you're going to get the actual character doing any kind of acting, that's when things look fake.

— Alex Arfaoui, filmmaker