How do you blend AI-generated shots with live-action footage seamlessly?

Blend AI shots with live-action by anchoring every generation to your practical footage, then finishing both in one grade. The working methods:

1. Upload stills from your shoot as references

2. Keep AI on wide environment shots

3. Build the world outward from a locked practical shot

4. Match continuity cues across the cut

5. Finish the blend in post — grade, sound, speed ramps

1. Upload stills from your own shoot as references. Before generating anything, screenshot frames from your practical footage and load them into the invideo agent as visual references — the invideo agent is a conversational video creation tool with the current models (Veo, Kling, Seedance 2.0) available in one place, so your references and every iteration live in the same thread. Reference-anchored generations inherit the color, contrast, and look of your real material, which is most of what "seamless" means. "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," says Alex Arfaoui, the filmmaker behind one documented hybrid production. A side effect: once the first shot in a scene establishes the look, tone, and effect, later shots in that scene land faster — that same production got a correct result on the very first generation of its second cave-wall shot.

2. Keep AI on wide environment shots — keep close-ups practical. Close framing on an AI character doing any acting is where the illusion breaks: "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." Pull out rather than push in. Shoot faces, hands, and performance practically, and hand AI the wides — in the documented production, the actor could tread cold water for only one to two minutes per take, so the tight emergence shots were filmed practically while AI supplied the wide ocean around them.

3. Build the world outward from a locked practical shot. Treat your practical footage as the anchor and ask the invideo agent to generate wide environment shots that place the same action in a larger world. That production's pool was a 6x6-foot studio tank — too small to hide set edges — so AI built the red ocean and shoreline wides around it; a separate 8x8-foot cave wall was physically constructed, with AI extending effects on top. Let story logic pick which environments to generate: the shoreline wide worked because the practical screenshot showed the actor on hands and knees coming out of water, so the two shots read as one continuous scene. Practical sets still get built — AI extends them, it doesn't replace them.

4. Match continuity cues across every AI-to-practical cut. Check four things between adjacent shots: lighting direction, color temperature, contrast, and atmospheric elements — matching haze between an exterior and interior shot is the kind of cue that makes a transition believable. When a generation nails most of these, pin it and request variations — "create another one like it, but change X, Y, and Z" — rather than starting over, or use invideo's extend to continue a shot from its final frame so consecutive AI clips share the same frame data.

5. Finish the blend in post — grade, sound, and speed. Color grade AI and practical footage to one look (the documented production used Lumetri Color in Premiere Pro), and layer sound design over every AI insert — sound is the multiplier that makes generated visuals feel physical and real. Two salvage moves for clips that still read as artificial: speed ramp them 1,000–5,000% and cut the fragments together as fast, chaotic inserts, and splice the best seconds from multiple imperfect generations into one polished sequence — first-generation output is rarely final.

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

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