Plan a hybrid AI and live action shoot by splitting the shot list before you book anything. Film what a viewer will inspect closely: faces, speech, hands, and the product with its label. Generate what is costly or impossible to film: distant locations, seasons, backstory and transitions. Then film everything so the generated shots can match it.
What a hybrid AI and live-action shoot is
A hybrid shoot is one video made from two sources: camera footage for what must be real, and AI-generated footage for what is impractical to film, cut together so the viewer sees one piece.
If you have not written prompts for video before, start with the AI video prompt examples and how to write your own before you mark any shot as generated.
What to film and what to generate
Film a shot when a mistake in it would cost you trust or a legal problem. Generate a shot when filming it would cost a travel day, a location fee or a season you do not have.
Shot type | Choice | Reason |
|---|---|---|
People speaking to camera | Film | Lip sync, eye contact and a real voice are what viewers judge hardest |
Product close-ups, labels, claims | Film | Logos, small print and exact colors must be correct in every frame |
Hands using the product | Film | Pouring, opening and pressing depend on real physics and real fingers |
Wide establishing shots | Generate | No face or label to inspect, and a real location costs a travel day |
Seasons, weather, time of day | Generate | You cannot schedule snow or a sunrise around one shoot day |
Backstory and "imagine if" scenes | Generate | Viewers read them as illustration, so a stylised look is acceptable |
Transitions between scenes | Generate from a filmed frame | The clip starts on your real footage, so the cut is hidden |
Customer testimonials | Film | A generated customer is a fabricated endorsement |
When a shot sits between the two, film it.
A shot-by-shot example
Take a 30-second launch video for an insulated water bottle, made by a founder, one camera operator and one actor in a single day. Five shots are filmed, three are generated. It is a planning example, not a finished production.
Shot | What the viewer sees | Source | Why |
|---|---|---|---|
1 | Dawn over a mountain trail, slow push in | Generated | Sets the promise; nobody inspects a landscape |
2 | Actor fills the bottle at a kitchen tap | Filmed | Water, hands and the product in use |
3 | Bottle on the counter, the lid clicks shut, logo sharp | Filmed | Hero product shot, the label must be exact |
4 | Bottle on the counter; the kitchen dissolves into a snowy ridge | Generated from the last frame of shot 3 | The transition starts on real pixels |
5 | Wide winter ridge, a small figure walking | Generated | No winter, no mountain, no budget for either |
6 | Actor drinks, breath visible, tight on face | Filmed at a cold doorway at dawn | A real face and a real reaction |
7 | Actor says the one-line benefit to camera | Filmed | Speech and eye contact |
8 | Product on a plain surface with the end card | Filmed, graphics added in the edit | Type and logo come from design files, never from a video model |
Export the final frame of shot 3 as a still and give it to an image-to-video model as the opening frame, so the generated clip begins on your own footage.
Prompt for the shot 4 transition
Locked-off camera, no camera movement. The steel water bottle stays exactly where it is, unchanged, with its label unchanged. The kitchen counter and wall behind it slowly dissolve into a snow-covered mountain ridge at dawn. Cold blue light replaces the warm kitchen light over four seconds. Fine snow drifts across the frame. No people, no text.
"Locked-off" matches the tripod shot before it. Naming the bottle and label as unchanged tells the model what not to redraw. "No text" is there because video models tend to invent lettering. If the label drifts in any frame, lay the bottle from your filmed frame over it in the edit.
How to film so AI footage matches
Generated shots match when you decide the look first and capture reference for it on the day. Viewers notice a cut when light direction, lens feel or motion changes.
Decide before the shoot
Fix one aspect ratio and one frame rate for the whole piece. Then choose the lens feel (wide and deep, or long with a soft background), the position and warmth of the key light, and the camera movement. Write all of them into every prompt.
Capture on the day
A clean plate of every setup. A few seconds of the empty frame, the cleanest starting image for a transition.
Product stills from several angles on a plain background, in the same light as the hero shot.
Wardrobe and face reference stills of the actor, if any generated shot shows a figure.
A color chart at the start of each setup. Visual effects crews use one to set exposure and white balance consistently across footage, as CAVE Academy's on-set reference notes describe.
Hold the first and last frame. Keep the camera still for two seconds at each end of a take, so you have a sharp frame to hand to the model.
Test the cut before the shoot day. Film ten seconds on a phone in the real location, generate one matching clip, and view them side by side in the delivery format. Mismatches show while you can still change the plan.
The post-production workflow
Edit the filmed footage first, lock it, then generate to fit the gaps. Generate earlier and you pay for clips of the wrong length.
Generating waits for two things from the filmed side: the locked edit, which fixes each shot's length, and the frames and stills you captured on the day.
Rough cut with placeholders. Cut the filmed shots and drop in a card, with its length, where each generated shot goes. The editor signs off the timing.
Edit lock. The story approver signs off. Shot lengths no longer change.
Generate. Make several takes of each shot from the frames and stills you captured.
Composite and clean up. Add logos and type from design files.
Color. Grade filmed and generated shots together, then add one layer of grain across both.
Sound. One music bed and one ambience track under every shot. Continuous sound hides joins.
Continuity and legal review. The brand owner checks every frame of the product.
For the generating step, Protoface's video models accept a still as the opening frame of a clip, and some models also accept a closing frame or reference images. Clip length is set per model, so check its catalog entry before you promise a shot length. The Protoface workflows include image to video, where you upload one image and describe the motion, and a video extender that continues a clip you upload.
Where an AI presenter fits in a hybrid shoot
An AI presenter, a generated version of a real person who speaks new lines, fits after the shoot day: for pickups, language versions and updates. It is a poor substitute for the main performance.
Pickups. A corrected line when you cannot recall the actor.
Language versions. The same script, with the mouth matched to new audio.
Updates. A new offer or date, months after the shoot.
To make the match possible, film a clean reference of the presenter on the day: facing the camera, evenly lit, same wardrobe, same framing as the main take, and a minute of natural speech recorded on the same microphone. Protoface's avatars do a different job: a live face for a voice agent in a realtime conversation.
Consent comes before the reference take. Do not record presenter reference "just in case". Agree in writing what the generated version may say, where and for how long before the camera rolls.
Costs, rights and consent
A hybrid shoot moves cost from the shoot day to post-production. It removes travel, location fees and extra crew days, and adds generation credits, rejected takes and longer review.
Every take is metered, including the rejected ones, some models by output length and others by the request. The article on AI video generator pricing, quotas, credits and budgets explains how to estimate a project. If the piece is a pitch and not a finished ad, a fully generated concept trailer may be cheaper still.
Likeness and consent
Get written consent before you create or alter footage of a real person. The US Copyright Office's report on digital replicas uses the term for a video, image or audio recording that has been digitally created or manipulated to realistically but falsely depict an individual, authorized or not. A generated pickup of your actor fits that description, so make it an authorized one. Likeness rules differ by country and by US state, so have a lawyer review the release. It should name the uses, the territories, the term, whether new lines can be generated, and who approves them. If your actor is a union member, ask their agent which contract terms apply before the shoot.
Who owns the generated shots
In the United States, expect your filmed footage to be on firmer ground than your generated shots. The Copyright Office's report on copyrightability, published in January 2025, concludes that copyright does not extend to purely AI-generated material, and that prompts do not alone provide sufficient control with the technology generally available at the time. It also concludes that copyright protects the original expression in a work created by a human author even if the work includes AI-generated material, and that whether a human contribution is enough is decided case by case. That is the Office's reading of US law, not a ruling on your project. Keep the moments you most need to own on camera, and ask a lawyer before you rely on owning a generated shot.
Disclosure
Realistic generated scenes may need a label. YouTube's rules on disclosing AI-generated content require creators to disclose when they use AI to meaningfully alter or generate photorealistic content: content that makes a real person appear to say or do something they did not do, alters footage of a real event or place, or generates a realistic scene that did not actually occur. Clearly unrealistic content and minor edits are exempt. In the EU, Article 50 of the AI Act says deployers of an AI system that generates or manipulates content constituting a deep fake shall disclose that the content has been artificially generated or manipulated. For an evidently artistic, creative, satirical or fictional work, the duty is limited to disclosing that such content exists, in a way that does not hamper the display or enjoyment of the work. Check the ad policy of each platform you buy media on.
Hybrid shoot planning checklist
Work through the three stages in order, because each protects the next.
Pre-production
Mark every shot in the shot list as filmed or generated, with a reason.
Fix aspect ratio, frame rate, lens feel and light direction for the whole piece.
Run a one-clip match test and budget for rejected takes.
Sign likeness releases that cover any generated use.
Shoot day
Clean plate and color chart for every setup.
A held frame at each end of every take.
Product and wardrobe stills in the hero light.
Presenter reference take, only if the release covers it.
Post-production
Lock the edit before generating.
Grade and grain filmed and generated shots together.
Frame-by-frame product check, then disclosure labels where platforms require them.
Common questions
What is a hybrid AI approach?
In video production, a hybrid AI approach means filming the shots that must be real, such as people, speech and the product, and generating the shots that are impractical to film, such as distant locations or seasons. Both are cut, graded and mixed as one piece.
What is the difference between AI-native and hybrid AI filmmaking?
An AI-native film builds its visual world primarily through AI systems, without depending on a full physical shoot. A hybrid film combines live-action footage with AI-generated visuals inside one directed workflow, as Saganix Studios' production notes put it.
What is a digital twin in video production?
The term is used for two things: a spatially accurate digital copy of a real location, used for planning and backgrounds, and a generated likeness of a real person. The US Copyright Office calls the second a digital replica. Get that person's written consent before you make one.
Does a hybrid AI shoot cost less than a full live-action shoot?
It usually cuts travel, location and extra crew days, and adds generation, review and continuity time in post-production. The saving is largest when the generated shots replace locations, and smallest when they replace shots you could have filmed in the same room.
Do I need permission from actors to use AI versions of them?
Treat it as required. Get written consent before the shoot that names the uses, territories, term and who approves any newly generated lines. Likeness law differs by country and by US state, so have a lawyer review the release. If the actor is a union member, ask their agent which contract terms apply.
Generate the shots you cannot film
Upload a frame from your footage, describe the motion, and get a clip that starts on your own shot. Protoface workflows cover image to video and extending a clip.
