Few industries have been given more confident predictions about their imminent replacement than animation and visual effects. Few have quietly absorbed the technology more pragmatically. Two years on, the studios using generative imagery seriously are not the ones that replaced artists with prompts, they are the ones that identified the specific, unglamorous stages where a generated image saves hours and does not touch the frame that ships.
This is a look at where it has genuinely landed in production pipelines, and where studios have drawn a firm line.
Pre-production is where it won
The clearest adoption is upstream of anything a viewer will ever see. Concept exploration, mood boards, colour scripts, environment ideation, lookdev references, the visual thinking that happens before a shot exists.
The reason is structural. Pre-production is iterative by nature, its output is disposable by design, and its historical bottleneck was never talent but throughput. A concept artist who could previously develop three environment directions in a day can now put twenty in front of a director, then hand-develop the one that survives. The generated frames are not the deliverable. They are the argument.
Art directors describe the change in practical terms: fewer meetings spent describing an idea, more spent evaluating one. When everyone in the room is looking at the same version of overcast, low sun, wet asphalt, muted palette, the discussion is about whether it serves the story rather than about whether everyone pictured the same thing.
Texture, matte, and set extension
Further down the pipeline, adoption is narrower but real. Generated source material has become common for texture bases that will be heavily processed anyway, for background matte elements far from focus, and for set extension plates where the requirement is plausibility rather than specificity.
The constraint here is technical rather than philosophical. Generated output arrives without the layer structure, camera metadata, or consistency guarantees that downstream stages assume. It works where the artefact will be substantially reworked and fails where it must integrate exactly. Studios that treat it as a starting material rather than a finished asset get value; those that expected a plate they could drop in did not.
The cost reality
The economics have shifted more than most coverage acknowledges. Generation is billed per image at fractions of a cent to a few tens of cents depending on resolution and quality tier, with no commitment. For a studio exploring a hundred concept directions in a week, the raw spend is negligible against a single artist-day.
Teams comparing options can review published rates directly rather than relying on vendor claims. Rates for the Nano Banana API and competing image models sit openly on platforms exposing several models through one account, which matters because the model best at photoreal environments is rarely the one best at stylised or graphic work, and a studio doing both is asking two different questions.
The figure that actually governs the bill appears on no pricing page: how many attempts precede an accepted result. Real practice runs three to eight, meaning true cost per usable frame is several multiples of the headline rate. The habit that controls it is straightforward, generate exploratory passes at low resolution, select, then regenerate only the chosen direction at full quality. Studios that build this into their workflow rather than leaving it to individual discipline consistently report halving spend with no difference in what reaches the director.
Consistency remains the hard problem
One image is trivial. Forty images that look like they belong to the same production is where most studio experiments stall, and it is the part of this that is genuinely a craft problem rather than a tooling one.
The teams that solve it treat prompts as a style bible rather than as individual requests. A fixed vocabulary, the same descriptors for lighting, palette, lens character, material response, and level of stylisation is maintained centrally and reused verbatim, with only the subject changing. It reads as documentation work. It is also the entire difference between a coherent visual language and a folder of unrelated images that happen to share a project name.
The line studios do not cross
Every production that has avoided trouble has drawn the same boundary, and it is worth stating because it is not primarily an ethical position.
Generated imagery is used for exploration, reference, and disposable pre-visualisation. It is not used for final frames, not for character likeness, and not for anything that will be represented as the work of a named artist without that artist’s involvement. The reasons are contractual as much as creative, talent agreements, guild terms, and client contracts increasingly address this explicitly, and a studio that discovers the boundary during a dispute has discovered it too late.
A related operational requirement that studios adopt late and regret: keep a record of which assets were generated, with model and prompt attached. Clients, broadcasters, and festival submissions increasingly ask. Capturing it as you go costs nothing; reconstructing it from a shared drive eighteen months later is a project nobody has budget for.
What the technology did not change
The barrier to producing a competent image collapsed. The barrier to knowing which image the sequence needs did not move at all.
Every artist who has used these tools in production reports the same thing: the bottleneck moved from execution to direction. Generating fifty variations takes minutes. Recognising which one serves the story, matches the established language, and will survive contact with the rest of the pipeline still takes the judgement that took a career to build.
That is not reassurance. It is a description of where the remaining work sits and where studios investing in people rather than in prompt libraries are placing their bets.
A note on smaller studios
The adoption pattern differs by scale in a way worth naming. Large facilities have absorbed generative imagery into existing pre-production departments, where it accelerates work that was already happening. Smaller studios and independent teams report something different: it has made pitch material viable that previously was not attempted at all.
A three-person team bidding on a commercial can now arrive with a developed visual treatment rather than a verbal description and two reference photographs. That is not a cost saving, it is access to work that was previously out of reach because the pitch itself was unaffordable. Several small studios describe this as the single largest practical effect on their business, well ahead of anything happening inside the pipeline.
