Skip to content
← Latent

Craft/

iQIYI's new studio keeps the actor and changes everything else

China's answer to the performance problem, and the most APAC-relevant development of the year.

On 2 November 2026 iQIYI opened an AI film studio in Shanghai, described as China's first. It is roughly 110 square metres, which is smaller than most photography studios in the city. The size is the interesting part. This is not a volume stage or a virtual production facility. It is a capture room built around a single idea: shoot the actor's performance for real, then change everything else afterwards.

The method works like this. Live-action capture syncs to iQIYI's NadouPro platform, which has access to around 30 models, including Seedance 2.5 and Kling 4.0. The captured footage becomes the reference. Sets, lighting, costume, makeup and props can then be altered in post while the original performance survives intact. Liang ZHU, iQIYI VP and Head of Intelligent Production, stated the premise without hedging: "An actor's performance is the core of any film or series." The inaugural title, "Cop in Trouble", began filming on 28 October 2026. The announcement.

110

square metres, smaller than most photography studios in the city.

For a studio working in this region, this is the most relevant development of the year, and not because of the technology. The models are available to anyone. What iQIYI has done is commit capital to a specific answer about which part of a film should be generated and which part should not.

The performance problem, stated plainly

Generated video has got good at surfaces and stayed weak at people. The Ipsos and Syracuse study found AI weakest on storytelling and emotion. The study. Those two findings are the performance problem under a different name. A model can produce a face that photographs well and a delivery that lands nowhere.

Most attempts to solve this have tried to generate better performances. iQIYI's facility declines the problem instead. If the performance is captured from a human being, it does not need to be solved. What remains to be generated is environment and production design, which is the part generated video is genuinely strong at, and the part that costs the most to build physically.

“Capture the thing that has to be believed, generate the thing that has to be afforded.”

That reframing has consequences for how a production is planned. It makes the shoot day small and the actor central. It moves location, set build and wardrobe spend into post. And it makes the captured plate the most valuable asset in the pipeline, which means capture quality, coverage and continuity matter more than they would on a conventional shoot, not less.

The same pattern, three other places

iQIYI is the first to build a facility around it, but the pattern is already visible in finished work elsewhere in 2026.

  • Promise shot Harlan Coben on an LA volume stage and composited more than 500 AI shots behind him for CBS, premiered 5 October 2026. Nem Perez framed the brief as a reconstruction problem: "How do you use GenAI to reconstruct a crime scene from only a handful of archival images and make it feel real?" The reporting. A real presenter, generated world.
  • Knucklehead and Airhead shot the car for real on MG Motors' "The Pentathlete" and generated the pentathlete vignettes around it. The interview. The hero product stayed photographic. The context did not need to be.
  • Netflix used generative AI in roughly 300 titles in 2026, concentrated in post-production. The breakdown. Not generated shows. Generated work inside shot shows.

Three different scales, one shape. Capture the thing that has to be believed, generate the thing that has to be afforded. The verifiable shipped work of 2026 is overwhelmingly hybrid, and iQIYI has built a facility around that fact rather than around the promise of full generation. That is the part worth copying.

It also sits on the right side of the union terms

SAG-AFTRA's four-year contract, ratified 5 June 2026, permits producers to use synthetic performers only where they bring "significant additional value" over a live actor or a digital capture of one. The terms.

A pipeline built around capturing a real performance and changing the world behind it sits comfortably with that language. A pipeline built around replacing the performer runs straight into it. iQIYI is not working under a US union agreement, but any APAC studio that wants to sell work into US and European productions will be, directly or through a client's compliance process. The architecture of the pipeline becomes a commercial question, not only a creative one. A method that keeps the actor is easier to clear, easier to insure and easier to defend in a client review.

What it does not solve

Keeping the performance does not remove the consistency problem, it relocates it. If the set, costume and lighting are generated per shot, they now have to match across shots that were captured against the same plain room. Continuity moves from the art department to the reference set. On the evidence of character work elsewhere this year, that is real effort rather than a checkbox, and the facility's output on "Cop in Trouble" is the first chance to see how well it holds across a full series.

It also does not reduce the actor's day to a formality. A performance given in an empty room, to be placed into an environment that does not exist yet, is harder than a performance given on a dressed set. That is a known constraint from volume-stage work, and it argues for more rehearsal and clearer direction on the capture day, not less.

What to take from it

For brand work the translation is straightforward, and we are describing a method rather than claiming results. If a film needs a person the audience must believe, shoot them. Keep the capture clean and keep the plate. Then put the budget into the world around them, where generation is strongest and physical production is most expensive. Specify which elements are captured and which are generated at the brief stage, because that line is the whole production plan.

iQIYI's version of this is 110 square metres and one platform with 30 models behind it. The idea it encodes does not require either. It requires deciding, before the shoot, which part of the film is not up for generation.

sillyhumans is an AI film studio in Singapore, making brand films and campaigns for brands internationally.

See the workStart a brief

Next — Industry Who is actually shipping Read