I talked for almost three hours on Sunday. Voice gone. Worth it.
This is the long version of the recap I posted on LinkedIn. It has every film, every partner, the photos, and the parts of the day I keep thinking about.
Who showed up
195 people applied for a room that fits far fewer. I read every application before the event. All of them. Launch videos and ads. Product demos. A short film someone had carried in their head for years. Many people asking the same quiet question: which tool is worth paying for, and which one should I skip.
The room at Yes SF on Bush Street ended up with three kinds of people sitting next to each other. Experienced videographers who shoot for a living. Filmmakers who had never opened a video model. Engineers who had never cut a scene. And one team making a film for their grandpa.
That mix was the whole plan. Builders learn film language fast when a director is next to them. Filmmakers stop fearing the model when an engineer shows them it is just another camera with strange settings.
The masterclass
Event linl: https://luma.com/pw54q8np
The morning block was scheduled for two hours. It ran three. People stayed.
I taught the way I direct a shoot. Story first, then the camera, then everything else.
Directing a model instead of prompting it. Longer prompts produce longer slop. A shot list produces a film. We went through how a director thinks about a scene, then translated that into instructions a video model can follow: framing, lens, lighting, movement, one idea per shot.
Film language. Why a 35mm feels different from an 85mm. How a low angle changes who holds power in a frame. Why the cut lands on the action, and how to make the model give you the action.
Consistency. The problem everyone hits: the same character, the same product, the same style across 40 shots. We covered reference workflows, how to lock a look, and when to give up on generation and use a real photo.
Mixing generated and recorded footage. The films people remember blend both. I showed how I integrate AI scenes with camera footage from Bayhaus projects, and how to match color, grain and motion so the audience stops asking which is which.
Editing, sound, finish. Motion graphics, soundtrack, pacing a hook in the first two seconds. Sound design from ElevenLabs, upscaling with Magnific, 3D assets from Tripo for shots that need a real object in space.
Which tool for which job. This was the part people had asked about most in their applications, so I spent real time on it. Image models, video models, voice, 3D, upscaling. What each one is good at right now, and where it falls apart.
Featured partner sessions ran inside the block. Each partner showed their tool, then stayed for the afternoon to help teams use it.
The hackathon
Lunch came from Sabor Gaúcho Prime Brazilian Steakhouse, which is my way of bringing Curitiba to Bush Street. Teams formed over picanha.
Then four hours. Strangers became crews. One table argued about a hook for twenty minutes. Another rebuilt their entire story with thirty minutes left. A third was fixing lip-sync while their teammate cut the trailer. Partners walked the floor and unblocked people in real time.
Every team had credits from the partners to spend during the event and afterward, so the work continues past Sunday.
The screening
At 5:30 PM the lights went down and we screened 23 films. Made in hours. By people who met at lunch.
Judges scored each film on story, execution, originality and impact, with a bonus for how well teams used the partner tools. Two tracks: Creative for films, Developer for tools and interactive experiences.
Creative winner: In Her Language
By James Yin, Ph.D. and team. A story about language and family that landed emotionally in a room full of strangers. Highest story and originality scores of the day.
Special credit to Xiaoqing Lin for the storytelling work.
Developer winner: Prison Escape! An Interactive Video Experience
Watch the demo:
By Peter Martin. A branching interactive film where the audience chooses the escape route. Ranked first overall across all judges, with the top impact score of the event.
The nine-year-old
A nine-year-old presented his film to a room of 100+ adults and won a nominated award. He spoke clearly, explained his choices, and took questions. I have directed CEOs who handled a stage worse than that. If you want to know where AI filmmaking is going, that is the answer.
Both winning teams shipped something I would put in a client reel.
Photos
Liz Zhang from The Multimodal Society
Loc H. Nguyen, Ed.D. from Bayhaus Creative
Every film from the day
Ranked by the judges’ average where scores were complete. Watch them all. Remember each one was made in about four hours.
Developer track
Creative track
- In Her Language, winner
- Epic Rap Battles of the SEC
- They Generate. You Create.
- Time To Move On
- Gut Run
- The Lightkeeper
- The Chase
- The American Dream
- Ollie Explains
- Cupify
- Last Catch
- trees
- Ova product launch video
- Ambitious Mother Podcast
- Avatar: The Cancelled Movie
- He Ran for His Life… Or Did He?
- El Colmo
The judges
Three people who took this as seriously as the teams did, and stayed for the whole screening.
- Long Do, Member of Technical Staff, OpenAI
- Calissa Man, Multimodal Post-training, ByteDance
- Bryan Chellen, Principal, Alumni Ventures
- Liz Zhang
- Roan Weigert
The partners
Each partner put credits on the table and then put people on the floor.
- Magnific, with Enrique Lopez Lopez. Upscaling and detail that made hackathon footage look finished.
- Tripo AI, with Aiko Dai and Ren Fu. 3D generation for teams that needed a real object in a real space.
- ElevenLabs. Voice and sound design.
- GMI Cloud, with Yuqi Hou. Inference for image and video models, and the GMI MCP tools teams used inside their workflows.
The hosts
Liz Zhang and The Multimodal Society co-hosted and ran the room. Loc H. Nguyen, Ed.D. from Bayhaus Creative handled production. Yes SF gave us the space.
What I learned
Read the applications. Reading all 195 changed the masterclass. I cut a section on model history and added forty minutes on tool selection because that was what people asked for.
Mixed rooms work. The best teams had one person who could tell a story and one person who could make a tool behave. Homogeneous teams produced technically clean films with weaker stories, or strong stories that fell apart on execution.
Constraints help. Four hours forces choices. Teams that picked one idea and one look early finished with something watchable. Teams that kept options open ran out of time.
Screen the same night. The room during the screening had an energy I have rarely seen at industry events. People watched their own work land on a big screen with strangers reacting. That is the reward, and it costs the organizer one projector.
Kids are ready. See above.
What comes next
More dates soon, in San Francisco first. The full online course that came with the ticket is open to everyone who was in the room, and to the people on the waitlist who could not get a seat this time.
If you want to be in the next room, the fastest way is to reach me on LinkedIn.
The collision between builders and filmmakers is the point. Sunday proved it.
