Google Screens 11 AI Shorts From Fourth Flow Sessions
Google screened 11 AI-assisted short films from the fourth cohort of its Flow Sessions program at Manhattan's Metrograph theater in late July, Variety reported on August 4. The six-week program gave participating filmmakers unlimited access to Google's Flow video generator and mentorship while they developed short story ideas for the screening.
Google screened 11 AI-assisted short films from the fourth cohort of its Flow Sessions program at Manhattan's Metrograph theater in late July, Variety reported on August 4. The cohort took part in a six-week intensive with unlimited access to Google's Flow short-form video generator and several mentorship sessions while developing short story ideas.
Variety described the directors as a mix of ages, backgrounds, and professional experience. Google launched Flow Sessions in September 2025 as a pilot that paired filmmakers with product access, mentorship, and AI education. The company later said its first cohort used the program to explore how Flow could fit creative workflows.
Eleven films, several production approaches
The screening included projects that used generated material in different ways. Variety reported that Dion Lee appeared in her six-minute film Two Chairs, which combined her performance with AI-assisted animated sequences around a personal question about having children. Nothin' Beats a Cheeseburger used two live actors while AI supplied an alien's appearance, according to the report.
Other films relied more heavily on generated scenes. Variety described Daniel Garcia's Mascot as wholly AI-generated, including behind-the-scenes images shown in its credits. The outlet also reported that a visible Nintendo Game Boy logo in Santiago Carrasquilla's One Breath at a Time was blurred for the online release, illustrating how generated imagery can still create ordinary clearance and brand-review work.
The films do not establish a new Flow model, benchmark, or production-quality metric. They show how Google is putting its video tool into a time-bounded creator program and presenting the resulting work in a theatrical setting.
What the program does and does not show
Google's own account of the first Flow Sessions cohort emphasized a director's mindset, experimentation, and personal storytelling. Those are useful creative principles, but they are company-reported lessons rather than evidence that the tool reduces cost, accelerates production, or performs reliably at studio scale.
For creative-technology teams, the fourth-cohort screening offers a more concrete workflow signal: participants used combinations of live performance, generated imagery, animation, and post-production decisions rather than a single automated filmmaking pattern. That makes provenance, asset review, editability, and rights clearance relevant alongside visual quality.
The program is also an adoption effort. Variety connected it to Google's broader work with filmmakers and its DeepMind research partnership with A24. The public evidence supports a claim that Google is courting creative professionals through access and mentorship; it does not yet show how widely those workflows will be adopted or how they compare with conventional production on time, cost, or audience response.
Key Points
- 1Google's fourth Flow Sessions cohort produced 11 AI-assisted short films during a six-week program with tool access and mentorship.
- 2The Metrograph screening included hybrid live-action and generated work as well as wholly AI-generated scenes, according to Variety.
- 3The program demonstrates a creator-adoption format, but the retrieved evidence does not provide model benchmarks, production-cost savings, or studio-scale reliability metrics.
Scoring Rationale
The fourth-cohort screening documents a concrete creator program and several ways generative video was integrated into short-film production. Its relevance is practical but bounded because the evidence provides no model release, benchmark, cost comparison, or production-scale performance data.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
