Hybrid AI and Live-Action Production: What Should You Film and What Should You Generate?
“Can we do this with AI?” is too broad to guide a commercial. A useful production decision is made shot by shot: what must be true, what must repeat exactly, what would be difficult to film, and where a human performance carries the message. The answer may change within the same ten-second ad.
Decision rule: film evidence, material truth and nuanced performance when the campaign depends on them. Generate controlled impossibility, set extensions and version-heavy elements after the real-world anchors are approved.
Question 1: must the viewer trust this object or event as real?
Film the product when its exact shape, label, texture, interface or performance affects a purchase decision. Food, apparel fit, machinery, packaging and before-and-after claims are especially sensitive to visual substitution. A generated hero shot can be expressive, but it should not quietly change what the customer receives.
The same rule applies to places. If the campaign claims a real Vancouver location, service area or event, use current footage or clearly identify a conceptual visualization. Generated mountains behind an office window may be harmless fiction in a stylized ad; the same image can mislead when presented as the actual view from a property.
Question 2: must this element remain identical across versions?
Generation is good at variation and can be weak at exact repetition. If a product, logo placement, uniform, actor, room layout or prop must match across ten ads, create one source of truth:
- approved product and talent reference images;
- dimensions, colour targets and non-negotiable details;
- camera height, lens character and lighting direction;
- approved wardrobe, location and prop IDs;
- one keyframe for each recurring shot family.
A practical product workflow often starts with controlled photography. Capture the front, side, three-quarter view, material details and a lighting reference. These images can support compositing or generation, while the real product shot remains available for the moment where accuracy matters most.
Question 3: is the image physically difficult or only expensive by habit?
Generate when the concept requires a world, scale or transition that is impractical to build: a restaurant unfolding into a paper city, a product moving through impossible seasons, or a room changing architecture in one shot. Before choosing generation, simplify the action to one readable event and test it.
Do not use AI to avoid ordinary pre-production. A crowded two-person handoff, exact liquid physics or precise product transformation may require more iteration than a small practical shoot. Compare the cost of usable outputs, compositing and review, not the price of one generation credit.
Question 4: does the moment depend on human nuance?
A founder explaining a difficult decision, a customer describing an experience or an actor reacting across a long take may carry details that are hard to direct through generation. Film those moments, then use generated inserts, environments or transitions around them. A hybrid edit can keep the human centre without limiting the visual world.
If a real performer will be altered, document the approved reference, actions, voice use, media, territory and term. Consent for an ordinary portrait or voice recording should not be assumed to include a synthetic performance.
Build the edit before creating every version
Approve one master story cut before generating all crops and languages. Once the sequence, claims and performance are stable, build a delivery matrix.
- 16:9 master for website, YouTube or presentation use;
- 9:16 version with reframed action rather than a blind centre crop;
- 1:1 or 4:5 paid-social crops when required;
- clean version without captions or platform graphics;
- captioned and language-specific versions;
- source and disclosure record attached to the approved masters.
Plan Greater Vancouver filming as its own workstream
When a hybrid project includes a real location, confirm property control, access, noise, power, weather cover, parking and the effect on public space. The City of Vancouver notes that most filming and photography on City property and parks requires a permit, with different City and Park Board paths depending on the location. Richmond, Burnaby, Surrey and other municipalities have their own processes.
Moncepts can combine commercial photography, live-action video and AI creative production in Vancouver under one shot and delivery plan.
Frequently asked questions
Should the real product always appear?
If exact product appearance or performance affects the customer’s decision, keep an accurate filmed or photographed representation in the campaign. Generated interpretation can support it without replacing the proof.
Can AI replace a Vancouver location shoot?
It can replace or extend a fictional environment. It should not be used to present a generated location as a real business, property or event when that difference would matter to the viewer.
Which method is faster?
It depends on the shot. A simple location plate may be faster to film; an impossible transition may be faster to generate. Test the highest-risk shot and compare usable results, approvals and rework.
Sources and scope
Vancouver location and permit notes were checked against the City’s film permit instructions and filming overview on September 9, 2026. The shot decision map, source-of-truth fields and delivery matrix are Moncepts production conventions.