Amazon MGM Bets on a Pipeline While Two Indie Hits Outpace Star Wars

This episode of Denoised pulls together two threads from AI on the Lot: studios committing real money to AI production pipelines, and a box office weekend where a $750,000 horror film outgrossed The Mandalorian. We recap Amazon MGM's Gen AI Creators Fund and its new Project Nara platform, the online firestorm that pushed Jorge Gutierrez out of the program, the agentic tools Dreamina and Luma showed off, and what Obsession and Backrooms suggest about the YouTube generation of directors.

What does AI in production actually look like once studios start writing checks? This episode walks through Amazon MGM's first slate of AI-assisted shorts, the online backlash that drove a director out of the program, and a box office weekend that handed the chart to two YouTube-trained filmmakers working at micro budgets. The thread: studios and creators alike are betting on more shots at smaller budgets, and the tools and talent pipelines that get them there are still being assembled in public.

The AI on the Lot team scaled the event into three or four studio lots, including Studio 15 at Amazon, with around 2,000 people at peak and the whole thing sold out. Technical sessions stayed small enough to ask questions, and the hallway track was where most of the connections happened. We covered the prior event, and the contrast was less about scale than about who was showing up. Multiple people walked in cold, including a Netflix friend who told us he came specifically because he knew nothing about AI and wanted a place to start.

What we noticed: more projects are in production using AI. The skeptics-on-the-fence crowd has shifted to the figure-it-out crowd. People who left media and entertainment a decade ago for consumer tech jobs at Meta, Instagram, and TikTok are boomeranging back into AI roles inside studios.

Amazon MGM hosted AI on the Lot and opened with the Gen AI Creators Fund, an undisclosed pool backing three animated projects plus a live-action short. The slate:

We previously reported on Amazon MGM's AI production push when the closed beta was announced, and the studio's track record on hybrid AI work includes the 253 AI-generated shots in House of David Season 2. The Moses series with Ben Kingsley used a similar hybrid approach, combining LED volume cinematography with Gen AI backgrounds and motion transfer.

The Jorge Gutierrez fallout: Gutierrez, a well-known animator, drew significant online backlash for partnering with Amazon on an AI-assisted short. He ultimately exited the program. Reports surfaced of death threats and family doxxing.

Addy's take: Part of the heat came from Gutierrez having been publicly anti-AI before joining the project. Changing your stance after seeing the tools in action is allowed, and the nuance gets lost in the online pile-on.

Joey's take: The misconception driving the anger assumes AI animation means typing a prompt and getting a finished film. What's actually happening is using AI as in-between work alongside an animator, their team, and their developed characters, so a project that could not get funded at full traditional cost can still get made.

The structure of the fund: Amazon is treating these as proof-of-concept pilots and will greenlight what works into full projects. Animation is the starting point because the quality gap to a tier-two animated feature is much smaller than the gap to a live-action tentpole.

Amazon described Project Nara as a collaborative workspace that integrates AI production agents with tools creators already use: Maya, Blender, Nuke, Unreal, and the Adobe suite. The architecture is model-agnostic. AWS is officially a partner on the platform.

Addy's take: This reads as the evolution of AWS Nimble Studio, the cloud-based studio offering. Whatever Amazon MGM builds will lean heavily on AWS infrastructure.

The portability question: Studios guard IP closely and want AI running inside their own walls. A pipeline built for Amazon may not translate cleanly to other studios that want self-contained deployments. If Amazon ever spins Project Nara into a hosted product on AWS Bedrock or similar, the bespoke nature of studio pipelines makes a self-serve version a tough sell.

Two agentic workflow tools landed during the event. Dreamina Octo is ByteDance's chatbot-driven canvas agent on its consumer-facing platform, tied into Seedance for video generation. Octo handles ideation, character sheets, and shot breakdowns, then calls the underlying models to produce images and video. Seedance already supports multi-shot generation from a single prompt, so layering an agent on top opens up faster sequence-building.

Luma Agents takes a different angle. It is model-agnostic, calling out to whichever model fits the task rather than locking users into Luma's own stack. We covered the Luma Agents launch and its underlying Uni-1 unified intelligence model when it shipped. That decision lets Luma sell the agent as a product on its own merits rather than as a funnel for Dream Machine.

Where these agents help:

Where they fall short: Turning a spark into a fleshed-out script, finding the emotional tone of an act, or making genuinely original creative calls. An LLM does well as a knowledge base or a variation generator. Ideation that requires taste is still a human job.

Addy's take: "The more I look into the pre-generation part of the pipeline, the more I'm convinced that we're not gonna lose our jobs."

The common refrain from studio conversations at the event was not about cutting budgets. It was about spreading the same budget across more bets. Eighty percent of the people in the room described it the same way: spend $100 million, get ten films out of it instead of one.

The blackjack logic: A single $100 million tentpole is no longer a reliable hit. Ten $10 million projects give the studio ten shots at landing a breakout. If one hits $200 million, it pays for the others. The IP library a successful project produces is the long-tail revenue that justifies the bet.

The commercial side: Lots of hallway talk about commercials going fully AI-generated. The counter-take: as automated content saturates feeds, the appetite for handmade, bespoke work goes up. Super Bowl commercials still draw attention because they are recognizably built by humans. That dynamic likely sustains a market for craft commercials even as the volume tier gets automated.

The box office weekend turned the studio strategy conversation into a live experiment. Two indie horror films took the top of the chart:

Obsession was acquired by Blumhouse at the Toronto Film Festival for $14 million. Parsons did the viral Blender-animated Backrooms short in 2022, and as we previously reported, A24 signed him on the strength of that work. Parsons was 16 when the short broke, signed at 18, shot the film at 19, and the movie released when he was 20.

AMC Entertainment reported its highest May attendance since 2019. Three movies opening the same weekend drove people back to theaters regardless of which one they came for.

Joey's take: This is not the MrBeast-to-Amazon model where a creator builds an audience on a channel and then transfers it to a streamer. Parsons and Curry Barker (the director of Obsession) did not run channels with built-in followings that translated to ticket sales. They posted work on YouTube, got noticed, got signed, and made movies. YouTube is functioning as a replacement for the Sundance circuit of the '90s and 2000s. Cuarón, Soderbergh, and that generation came up on film festivals. This generation comes up by posting work where any exec can watch it.

Addy's take: A festival showcases work to maybe 500 people in the room. YouTube is a democracy. Any exec scouting talent can pull up the work and gauge quality without sitting through a Park City lineup.

Where AI fits in two years: Neither Backrooms nor Obsession used AI in any meaningful way. Their VFX needs were modest because they leaned on confined locations and practical builds. The next cohort of directors coming up the same pipeline will have AI in the pre-vis, storyboarding, and concepting stages, plus the ability to self-finance and self-build at higher fidelity.

The Star Wars footnote: Joey noted that Dave Filoni reportedly did not enjoy Andor, which was shot without virtual production, and Addy added that Andor was all physically shot. Addy called Andor a standout against the rest of recent Disney output, a contrast worth noting against the Mandalorian numbers.

The two halves of this episode connect on the same problem from opposite ends: a $100 million bet on a known IP is no longer a sure thing, and the next generation of directors is being discovered on the same platform their audiences already live on. AI is the tool both sides reach for to make more bets affordable.

The studios betting on volume and the creators emerging from YouTube are converging on the same answer: more shots on goal at lower budgets, with AI doing more of the work that used to require a full crew.

Events:

Studios & Programs:

Tools & Platforms:

Projects: