An AI entertainment studio · San Francisco, California
Characters, stories, and worlds. Built to grow.
We make characters people follow across text, image, voice, sound and video. Then we build the agents that find their audiences, run the experiments and learn what keeps people coming back.
One character, every format.
Text, images, voice, sound and video should all feel like the same character. We build every piece from one shared description of who they are.
Characters
Original personalities with a look, a voice, a history and a memory.
Stories and worlds
Settings with rules and timelines that can carry many stories.
Interactive experiences
Conversations and choices that change what happens next.
Short-form video
Scenes built to introduce a character in seconds.
Long-form storytelling
Episodes and arcs for the characters people come back for.
Mabel Quill
An overnight radio host whose callers ring in from other worlds.
- Look
- Lilac scarf, studio headphones
- Voice
- Warm, unhurried, low register
- Personality
- Curious, never rattled
- Memory
- Remembers every caller
- World
- Station K-LOW, 3 a.m.
- 01 · Text
“You're caller number nine tonight. The first eight were from places that don't exist.”
- Personality
- Memory
- World
- 02 · Image
- Look
- World
- 03 · Voice
Pauses before good news.
- Voice
- Personality
- 04 · Sound
Station chime: three notes, rising.
- World
- 05 · Short video
- 0:00 Phone lights up
- 0:04 Static clears
- 0:09 Caller says her name
- Look
- Voice
- World
- 06 · Long-form
Ep. 1 · The Caller Who Knew Her Name
Act two returns to caller nine from the short.
- Personality
- Memory
- World
Agents that learn what works.
Our growth systems research opportunities, make creative, run experiments and change their next decision based on what the last one earned.
- 1
Research
Agents scan search demand, communities and partner channels for audiences worth reaching.
- 2
Create
They draft briefs, landing pages, articles and creative variants for people to approve.
- 3
Experiment
They run controlled tests, with holdout groups where the question needs one.
- 4
Learn
They read the outcome in retained revenue and update the playbook.
- What they learn: which audiences, hooks and formats bring people who stay, which pages and messages turn interest into revenue, and when an idea has run its course. What changes: the briefs they write, where budget goes and which test runs next.
playbook.md · revision 2
Illustrative example- Hook
−Open with the premise.
+Open on the character mid-scene.
Learned from: Holdout groups shown the character first returned more often.
- Landing path
−Send every visitor to the same page.
+Match the page to the scene that brought them.
Learned from: Visitors from different scenes dropped off at different steps.
- Follow-up timing
−Follow up seven days after a first visit.
+Follow up when the next episode is released.
Learned from: Messages tied to new story beats were opened and acted on.
- Budget
−Split spend evenly across new concepts.
+Reserve a fixed share for exploration; scale what retains revenue.
Learned from: Concepts that won on clicks did not always win on revenue.
Every release is an experiment.
Creative decisions produce audience feedback. Feedback makes the next creative and growth decision better. The loop is the company.
- 1
Create
Writers and generation pipelines make characters, scenes and campaigns.
- 2
Distribute
Agents place the work in search, social and partner channels.
- 3
Measure
We track who stays and what they spend, against holdout groups.
- 4
Learn
Results update the playbook: which hooks, worlds and paths work.
- 5
Improve
The next round of creative and growth decisions starts from there.
Then back to Create, with a better starting point.
Open questions we are working on.
Each direction starts from a problem in production. Statuses say plainly what exists today and what is still a proposal.
Reinforcement learning for revenue growth
Can agents learn which sequence of creative, landing-page and lifecycle decisions produces revenue weeks later?
Agents that improve their own playbooks
Can agents inspect their own failures, revise their prompts and tool strategies, and beat their previous versions on tasks they have not seen?
Learning what makes creative work
Which character, hook, scene, pacing or visual treatment actually caused an improvement, and which just came along for the ride?
One character across every medium
How do appearance, voice, personality, memory and world history stay consistent across conversations, images, shorts and long-form video?
Built in production.
- 10M+
- People who have used products built by Playhouse Cumulative, since 2024
- 1T+
- Tokens generated across our systems Per month, 2026
- 2024
- Founded in San Francisco Playhouse Media LLC