What will movies be like without human actors?

On 20 April, Chinese online video platform iQIYI announced with great fanfare the establishment of its artificial intelligence (AI) talent pool. Its CEO, Gong Yu, also shared that his company is collaborating with GHY Culture & Media to develop a fully AI-generated movie for its Ferryman series. 

iQIYI is not the first to venture into AI-driven film and television, but in proposing to produce films generated entirely by AI, it may well be breaking new ground. The CEO offered two main reasons for AI-generated films: first, the trend towards AI in film and television is unstoppable; second, it can free actors from the burdens of intensive shooting schedules. Actors could license their likenesses to AI productions, gaining more rest time and bringing their working patterns closer to those of ordinary white-collar employees. Where previously they might take on two projects a year, they could now handle four. As a result, the development of fully AI films may one day render live-action filmmaking a kind of “intangible cultural heritage” — though it would not disappear entirely.

Protecting human actors

Developing, embracing and using AI has become a trend and a new mode of production and living. Film and television, by virtue of their reliance on images, sound and language, and their capacity to express emotion and the human spirit, are naturally suited to adopting and integrating AI. In this sense, the use of AI appears almost inevitable.  

However, even if the conditions seem ripe, it is premature to say that everything is in place. The key question is whether audiences are willing to watch — or even enjoy — fully AI-generated films, and whether the market can accept or embrace them. At present, there is no ready answer. Only time will tell, and ultimately audiences will decide with their wallets, subscriptions, remote controls and keyboards.   

While fully AI films could indeed ease the heavy workload of human actors, securing their consent is the first hurdle. iQIYI claims that its platform Nadou Pro has already secured AI licensing agreements with over a hundred celebrities and has established an “AI talent pool”, including Chen Zheyuan, Zeng Shunxi, Cheng Lei, Jiang Long, Ma Su, as well as stand-up comedian Director Fang (real name Fang Zhuren).

However, some actors have denied such involvement. Zhang Ruoyun, Yu Hewei, Li Yitong and others have issued statements through their studios or related parties, denying that they have signed any AI film and television authorisation agreements. This suggests that some actors remain wary of fully AI-produced films, or at the very least are keen to protect their reputations. They are unwilling to let AI replace their real-life personas, or to see the authentic images they have painstakingly built — through hard work, careful performance and sustained effort — in the minds and memories of audiences be replaced, deconstructed, blurred, distorted or even damaged by AI.

AI vs human actors

On the other hand, although fully AI films must ultimately be judged by audiences after they are made, some viewers have already expressed scepticism: if even live-action films struggle to attract audiences, who will watch AI films? If actors are all AI-generated, perhaps AI viewers should also be the ones paying for subscriptions — let AI both watch and pay for films.

Such remarks reflect several concerns. If actors contribute only their faces while all movements and special effects are handled by AI, the result may feel excessively artificial. Like plastic flowers — attractive from afar but lifeless, colourless and devoid of vitality up close — such productions may naturally spark resistance rather than liking.

There is already a precedent. Sora, in fact, was an early pioneer in AI-generated video. Developed by OpenAI, it is a text-to-video model capable of generating realistic, high-definition videos of up to 60 seconds based on text prompts.

However, in late March 2026, OpenAI announced the closure of Sora’s standalone app, API interface, as well as its integrated video-generation function within ChatGPT, marking its complete withdrawal from the consumer AI video market. The shutdown was attributed to several factors, including high costs, difficulties in commercial monetisation, and strategic downsizing amid pressure from an IPO. Among these, the difficulty of monetisation suggests that fully AI-generated video content has not been well received by the public and has failed to capture the market. This may foreshadow a similar fate for fully AI-generated films in the future.

Different quality

Text-to-video outputs are not uniformly poor, but a considerable portion lack appeal. A key issue is homogeneity: because the barrier to entry is extremely low, differences between works are greatly diluted. If fully AI production becomes widespread, anyone could sign agreements with actors — or actors themselves could generate films without personally taking part in filming — making it difficult to ensure quality.

There are also notable quality differences between AI-generated and real filmed footage. AI lacks a genuine understanding of causal relationships and physical logic — such as gravity, collision and occlusion — leading to unrealistic or illogical outputs. Examples include food passing through teeth during eating scenes, objects merging into one another, and inconsistencies in time and space. Within seconds, objects, characters or backgrounds may mutate: clothing colours may suddenly change, limbs may appear abnormal (extra fingers or missing arms). These issues stem from AI’s limited control over continuous frames. Textures also often lack realism: skin appears overly smooth, metals lack proper sheen, lighting fails to reflect environmental sources accurately, and depth is poorly rendered.

More critically, character movements tend to be stiff, producing the “uncanny valley” effect — when virtual humans or robots appear almost, but not quite, lifelike, triggering instinctive feelings of unease, discomfort or eeriness. AI-generated characters often exhibit unnatural expressions — vacant gazes, lack of dynamic muscle texture, rigid movements — making them instantly recognisable as “fake”, much like plastic flowers. If films are fully AI-generated in future, they are likely to suffer from the same issues, making it difficult to win audience approval.

Furthermore, fully AI films may invert the relationship between cause and effect. Only when actors create iconic roles through real performances and win audience approval do their images become popular; only then can AI replicate and utilise those images in subsequent productions. Without authentic performances first, there can be no meaningful AI generation later.

Whether fully AI films can secure a place in the market can only be determined by time, audiences, investment and box office performance. Even so, it may still be worth trying.

Related: Why Seedance beat Sora in the race for AI video generation | How Chinese talent powers US AI dominance