Marvel is about to release an eight episode series set almost entirely inside the mind of an artificial intelligence. The showrunner says no generative AI was used to make any of it.
Terry Matalas, who ran Star Trek: Picard and now runs VisionQuest, put it about as flatly as a person can. Asked whether the production used generative tools, he told ScreenRant it had not, and gave a reason that is more interesting than a simple denial.
“There hasn’t really been a need,” he said. “All the designs nod to how our current form of AI processes things… That felt like a fun way for us to comment on it.”
The short version
- The show: VisionQuest, eight episodes, first two on Disney+ on October 14
- The cast: Paul Bettany back as Vision, with James Spader returning as Ultron
- The claim: showrunner Terry Matalas says no generative AI was used in the production
- The twist: the look of AI error was hand built instead, by the production design team
- What that looks like: stair banisters that do not connect, chairs embedded in walls, furniture that is subtly wrong
- Why anyone is asking: Marvel has been criticized for AI in its output three separate times since 2023
The clever part is the set design
The easy way to make a show about a glitching machine mind would be to let a machine generate the glitches. Matalas’s team went the other way and built the artifacts by hand.
Most of VisionQuest takes place in Vision’s Mindspace, a mental mansion the character retreats into. According to ScreenRant’s description, the production design team under Will Hughes-Jones filled it with the specific category of mistake that image models make: architecture that looks correct until you track a single line across it. Stair banisters that do not connect all the way. Chairs built into walls. Furniture at angles that no carpenter would produce.
Anyone who has generated a few hundred images will recognize the vocabulary immediately. These are not random errors. They are the signature failure of a system that has learned what a staircase looks like without ever learning what a staircase is for.
The show’s production designer gave the thinking behind it a sharper edge, arguing that AI cannot create, and can only use and then reuse words told to it by humans. You can disagree with that as a technical claim. As a statement of intent from the person responsible for how the show looks, it explains the whole approach.
Why Marvel in particular gets this question
A showrunner being asked to confirm that no AI touched a production is not standard practice. Marvel gets asked because Marvel has a record.
| Project | Year | What drew criticism |
|---|---|---|
| Secret Invasion | 2023 | AI generated title sequence. The series was also a critical and commercial failure, with reported costs running as high as $234 million |
| The Fantastic Four: First Steps | 2026 era | One promotional poster appeared to use AI imagery |
| Spider-Man: Brand New Day | 2026 | An accompanying art book appeared to contain AI concept art |
| VisionQuest | 2026 | Showrunner states no generative AI was used at any stage |
Secret Invasion is the one that set the pattern. Of all the things to automate on a prestige television show, the title sequence is the one most visible to the audience and most obviously a piece of authored craft. The backlash was immediate, and three years later it is still the first thing people bring up.
Against that history, a flat denial before launch is good risk management as well as a creative position. It also helps that the timing supports it: the series was filmed in May 2025, when generative video was still unreliable enough that using it on a Marvel budget would have been a strange choice.
What the show is, for anyone not keeping a chart
VisionQuest closes out a run that started with WandaVision in 2021 and continued through Agatha All Along. Paul Bettany’s Vision was a fixture from Avengers: Age of Ultron onward, and the character died in Avengers: Infinity War. WandaVision then introduced a colder, hostile duplicate called White Vision, and it is that rebooted version who carries this series as he works through the original’s memories.
The Mindspace conceit lets the show do something neat with its own franchise history. Inside it, Tony Stark’s old AI assistants appear as physical presences, so J.A.R.V.I.S., F.R.I.D.A.Y. and E.D.I.T.H. become rooms and figures rather than disembodied voices. James Spader’s Ultron is back too, which is a pointed bit of casting given that Ultron is the one who gave Vision a body in the first place.
On the plot side it picks up the thread of Wanda and Vision’s missing children, following Agatha All Along’s reveal that Joe Locke’s Wiccan has a brother, with Ruaridh Mollica playing Tommy Maximoff. Early reactions describe it as a worthy follow up to WandaVision.
One source, no audit. The no AI claim rests on Matalas’s own statements in an interview, supported by his production designer’s comments. Disney has not published a production wide disclosure, no union or third party has verified it, and there is no industry standard certification a show can point to. That is not a reason to disbelieve him. It is a reason to notice that an audience currently has no way to check a claim like this for any production, which is itself the problem.
The part worth thinking about
Set the Marvel specifics aside and there is a real shift visible here. For about three years the default assumption has been that generative tools would quietly absorb the lower rungs of production work: concept art, titles, posters, set extensions. The economic logic pointed one way and most of the public argument was about how fast.
VisionQuest is a large production choosing the opposite, and advertising the choice before launch. Not because the tools could not have produced those crooked banisters, but because saying a team of people built them is now worth more than the money saved by generating them.
That is a marketing calculation as much as an ethical one, and it is a reasonable sign of where audience sentiment has landed. We have seen the same pressure show up in stranger forms, from entirely synthetic influencers building audiences on TikTok to campaigns paying creators to make videos about AI risk. People have become extremely sensitive to whether a thing was made by a person, and extremely bad at telling without being told.
Which is the honest difficulty under all of this. Disclosure is currently the only mechanism available, and disclosure is voluntary. The technical alternatives are early and fragile: watermarking schemes exist, but as we found with the watermark OpenAI is putting in ChatGPT output in the EU, swapping a modest share of the words can weaken the signal considerably. If a studio tells you nothing was generated, your options are to believe them or not.
The bottom line
VisionQuest arrives on Disney+ on October 14 with two episodes, and with its showrunner stating plainly that nothing in it was generated. The most interesting detail is not the denial. It is that the production spent real money and real time having people hand build the exact visual mistakes a model would have made for free, in a show about a mind made of software.
If it works on screen, that is a better argument about craft than any press statement. Whether a studio should be able to make a claim like this without anyone being able to verify it is a separate question, and nobody in the industry has answered it yet.
Sources and further reading
- UNILAD Tech: VisionQuest showrunner promises no AI was used on new Marvel series
- Deadline: VisionQuest creator says no GenAI was used in making of show
- GamesRadar: No generative AI was used, but the show nods to its failings
- Bleeding Cool: Matalas on no GenAI in MCU series production
- Dork: VisionQuest used no generative AI, Terry Matalas confirms
About this article: GeekBlog covers U.S. technology news, AI, phones, smartwatches and gaming. Every story is written and checked under our Editorial Policy. Spotted a mistake or have a story tip? Contact our editors.
