Matcha, code, and AI tools | Ex @Microsoft | Turning AI into real growth
Joined July 2026
- Tweets408
- Following93
- Followers12K
- Likes1.4K
ACTION MOVIES ARE AI NOW
boxing robots, a sword fight in a temple, a burning sea, a hallway shootout. all Kling 4.0
no stunt team, no set, no second unit
AI series that pull numbers right now: mafia romance, BL/GL romance and sci-fi survival. ARK-7, a sci-fi serial, has 1.97M views on episode 1 on a channel with 34.7K subscribers
pick a lane people already binge. the rest is in the guide below
NETFLIX'S NEXT COMPETITOR IS A WORLD YOU CAN STEER
this is Genie 4 from Google DeepMind. one prompt for the world and the character, then you walk through it and type changes while it runs. native audio
> nothing here was shot or rendered by a studio
> you change the scene live
> Google calls it its most realistic and interactive world model yet
stories are going interactive. the people who already know how to write a season get there first
how to write one, step by step, is below
2023: WILL SMITH EATING SPAGHETTI WAS THE PROOF AI VIDEO WAS A JOKE. 2026: THIS
This side by side has 3.5M views in a day.
> 2023: the face melts into the pasta and the fork goes through his chin
> 2026: a rooftop dinner, two actors, real light, cuts like a film
The same three years took the cost of a series down to almost nothing:
> Netflix reportedly spent $50-60M on each episode of Stranger Things 5
> a 20-episode AI series runs about $300 in credits
> about $12 per minute on screen
How to direct a whole season yourself, from logline to money, is in my article.
an anonymous TikTok account turned fruit into Love Island contestants
> episode 1: 36.6M views
> episode 15: 40M
> 2.1M followers
> nobody on screen was real
TikTok paid them about $15 per million views. ~$4,500 for ~300M views. on April 1 the series got pulled from the Creator Rewards Program
season 2 opened at 2.3M and slid to ~200K an episode within a month
so the views were free and the ad money was close to zero. the team moved to Telegram and brand deals
where AI series money comes from, platform by platform, is below
FRUIT LOVE ISLAND GOT 40M VIEWS ON ONE EPISODE. TIKTOK PAID ABOUT $15 PER MILLION
> episode 1: 36.6M views
> episode 15: 40M
> ~300M views earned about $4,500
> April 1: removed from the Creator Rewards Program
> season 2 opened at 2.3M and slid to about 200K an episode within a month
Novelty carried season one until it ran out. The team moved to Telegram and brand partnerships.
Where AI series money comes from, platform by platform, is in my article.
SOFIA retweeted
THIS DIRECTOR WENT FROM ABOUT 70,000 YUAN A MONTH TO 24,000 AFTER HIS WORK MOVED TO AI
Wang Jinfei ran film sets until last year. Now he directs AI characters in China's microdramas.
> investors used to spend 3 million yuan on ten productions
> he says it's now 300,000 for ten AI ones
> some companies say they can make AI content for $30 a minute
> people tell him he's lucky, because so many in film have no work at all
My read: Wang gets paid out of a budget that keeps shrinking. The person who owns the series keeps the upside.
This article shows how to run that pipeline yourself from Claude, from series bible to final cut, with a price on every batch before you pay.
FRUIT LOVE ISLAND GOT 40M VIEWS ON ONE EPISODE. TIKTOK PAID ABOUT $15 PER MILLION
> episode 1: 36.6M views
> episode 15: 40M
> ~300M views earned about $4,500
> April 1: removed from the Creator Rewards Program
> season 2 opened at 2.3M and slid to about 200K an episode within a month
Novelty carried season one until it ran out. The team moved to Telegram and brand partnerships.
Where AI series money comes from, platform by platform, is in my article.
TO SEE THE KISS IN THIS EPISODE YOU PAY 18 COINS, ABOUT 20 CENTS
That paywall helped ReelShort pull in about $1.2B in consumer spending in 2025, per Appfigures.
> a typical series: 60 to 90 episodes
> shot in 7 to 10 days
> up to $300,000 to make, per WSJ
> the whole market: $14B by the end of 2026, per Omdia
The same kind of series can now start with one line in Claude Code:
/picsart:make-short-drama a revenge series set on a hedge fund trading floor at night
Claude asks the showrunner questions first. Then it writes a series bible, locks the cast and quotes a price before any video gets generated.
All eight commands and the install are in this article.
THIS DIRECTOR WENT FROM ABOUT 70,000 YUAN A MONTH TO 24,000 AFTER HIS WORK MOVED TO AI
Wang Jinfei ran film sets until last year. Now he directs AI characters in China's microdramas.
> investors used to spend 3 million yuan on ten productions
> he says it's now 300,000 for ten AI ones
> some companies say they can make AI content for $30 a minute
> people tell him he's lucky, because so many in film have no work at all
My read: Wang gets paid out of a budget that keeps shrinking. The person who owns the series keeps the upside.
This article shows how to run that pipeline yourself from Claude, from series bible to final cut, with a price on every batch before you pay.
THIS DIRECTOR WENT FROM ABOUT 70,000 YUAN A MONTH TO 24,000 AFTER HIS WORK MOVED TO AI
Wang Jinfei ran film sets until last year. Now he directs AI characters in China's microdramas.
> investors used to spend 3 million yuan on ten productions
> he says it's now 300,000 for ten AI ones
> some companies say they can make AI content for $30 a minute
> people tell him he's lucky, because so many in film have no work at all
My read: Wang gets paid out of a budget that keeps shrinking. The person who owns the series keeps the upside.
This article shows how to run that pipeline yourself from Claude, from series bible to final cut, with a price on every batch before you pay.
SOFIA retweeted
NETFLIX IS DEAD
> Stranger Things 5: $50-60M per episode, reportedly
> Squid Game: about $44,000 per minute on screen
> a 20-episode AI series: about $300 in credits
> that's about $12 per minute
$300 is less than Netflix spent on one second of Stranger Things 5
Netflix knows it. it used generative AI for a building collapse in El Eternauta and finished the shot 10x faster, and in March it opened an AI animation unit
Netflix pays for a whole season before anyone watches. you can pilot 3 concepts for ~$100, post them, and fund the one people ask for
the full playbook, with a sample series built step by step, is below
SOFIA retweeted
HOW TO BUILD A 75-SECOND AI EPISODE
TikTok's Creator Rewards only pays for videos over one minute, and 75 seconds gives you margin. This clip is exactly that long.
> 0-5 s: the hook. Conflict in the first image, readable on mute
> 5-54 s: friction. One scene, one location, mostly faces
> ~54 s: the turn. The hero makes a choice
> last 5 s: the cliffhanger. The world does something to them
Filmustage's rule for the ending: "Cut on the question, not the answer."
Save this. The whole season plan, from logline to money, is in my article.
NETFLIX ALREADY USES GENERATIVE AI. IT DID THE BUILDING COLLAPSE IN EL ETERNAUTA 10X FASTER
Ted Sarandos said the traditional way "just wouldn't have been feasible for a show in that budget."
> in March it launched an AI animation unit, INKubator
> it plans to spend $20B on content in 2026
> this clip is the VFX breakdown for that show: built sets, blue screens, a full crew
A studio still pays writers, a set and actors. You need a laptop, a subscription and taste.
The solo version, step by step, is in my article.
NETFLIX IS DEAD
> Stranger Things 5: $50-60M per episode, reportedly
> Squid Game: about $44,000 per minute on screen
> a 20-episode AI series: about $300 in credits
> that's about $12 per minute
$300 is less than Netflix spent on one second of Stranger Things 5
Netflix knows it. it used generative AI for a building collapse in El Eternauta and finished the shot 10x faster, and in March it opened an AI animation unit
Netflix pays for a whole season before anyone watches. you can pilot 3 concepts for ~$100, post them, and fund the one people ask for
the full playbook, with a sample series built step by step, is below
HOW TO BUILD A 75-SECOND AI EPISODE
TikTok's Creator Rewards only pays for videos over one minute, and 75 seconds gives you margin. This clip is exactly that long.
> 0-5 s: the hook. Conflict in the first image, readable on mute
> 5-54 s: friction. One scene, one location, mostly faces
> ~54 s: the turn. The hero makes a choice
> last 5 s: the cliffhanger. The world does something to them
Filmustage's rule for the ending: "Cut on the question, not the answer."
Save this. The whole season plan, from logline to money, is in my article.
SOFIA retweeted
NETFLIX SPENDS $50M ON ONE EPISODE. THIS ONE COST $28,775
it's an episode of Nexus, an AI series. a guy with zero brain cells rides a dinosaur that shoots exploding pimples
> $28,775 to make
> Nexus episodes are at 62,479,089 views
> that's about 0.05% of one Stranger Things 5 episode
how to direct your own, for ~$300, is below
NETFLIX SPENDS $50M ON ONE EPISODE. THIS ONE COST $28,775
it's an episode of Nexus, an AI series. a guy with zero brain cells rides a dinosaur that shoots exploding pimples
> $28,775 to make
> Nexus episodes are at 62,479,089 views
> that's about 0.05% of one Stranger Things 5 episode
how to direct your own, for ~$300, is below
SOFIA retweeted
no os miento cuando os digo que quizá nos encontramos ante la mayor cuenta de Twitter de la historia relacionada con la IA, no he visto un tweet malo de esta cuenta
ESTO ES UNA MALDITA LOCURA
13 repos gratis que convierten una carpeta de notas en un segundo cerebro que un agente puede usar de verdad: captura lo que lees, busca en todo lo que has guardado y nunca empieza de cero
[01] capturar
obsidian-clipper: el web clipper oficial de Obsidian, guarda cualquier página como Markdown github.com/obsidianmd/obsidi…
markitdown: convierte PDF, Word, PowerPoint, Excel y audio en Markdown github.com/microsoft/markitd…
yt-dlp: saca los subtítulos de los vídeos sin descargarlos github.com/yt-dlp/yt-dlp
paperless-ngx: OCR y búsqueda de texto completo en todos tus escaneos y PDFs github.com/paperless-ngx/pap…
[02] estructurar
obsidian-dataview: convierte tu vault en una base de datos que puedes consultar github.com/blacksmithgu/obsi…
obsidian-git: historial de versiones y copia de seguridad automática de cada nota github.com/Vinzent03/obsidia…
[03] buscar
qmd: búsqueda local en tus notas, combina palabras clave, vectores y re-ranking, y tiene servidor MCP github.com/tobi/qmd
meilisearch: búsqueda mientras escribes, tolerante a erratas y con modo semántico híbrido github.com/meilisearch/meili…
[04] memoria del agente
letta: agentes con estado que aprenden con el tiempo github.com/letta-ai/letta
langmem: hechos, comportamiento e interacciones pasadas para la memoria a largo plazo del agente github.com/langchain-ai/lang…
MemOS: una capa de memoria para agentes, guardada como un grafo que puedes inspeccionar y editar github.com/MemTensor/MemOS
[05] que sea tuyo
quartz: publica tus notas en Markdown como una web github.com/jackyzha0/quartz
syncthing: sincroniza tu vault entre dispositivos, P2P y cifrado github.com/syncthing/syncthi…
Un montón de notas solo se convierte en un segundo cerebro cuando un agente puede buscar en ellas y recordarlas.
Todos los repos de arriba son gratis, y casi todo funciona en tu propio ordenador.
SOFIA retweeted
Google says Gemini 4 Argon built and edited this whole video. It scripted the Blender boot animations, wrote the SuperCollider audio, and the signal runs through a Raspberry Pi into a real CRT TV.
Same model in October:
> built a full Japanese pagoda in Blender with agents
> one agent builds the scene, another critiques the render, they loop until it holds up
> that demo has 15.8M views on YouTube
A year ago AI video meant a prompt box and a 5 second clip. Now an agent opens Blender and edits the timeline itself.
Google hasn't said how long any of this took or what it cost, so I want to see someone outside Google reproduce it.
Our team used Gemini 4 Argon to build and edit this entire video, then displayed it on a real vintage CRT TV 📺
We used Argon in @Antigravity to script the 3D Blender boot animations and retro SuperCollider audio, then piped the signal via Raspberry Pi to a real TV screen.
Google says Gemini 4 Argon built and edited this whole video. It scripted the Blender boot animations, wrote the SuperCollider audio, and the signal runs through a Raspberry Pi into a real CRT TV.
Same model in October:
> built a full Japanese pagoda in Blender with agents
> one agent builds the scene, another critiques the render, they loop until it holds up
> that demo has 15.8M views on YouTube
A year ago AI video meant a prompt box and a 5 second clip. Now an agent opens Blender and edits the timeline itself.
Google hasn't said how long any of this took or what it cost, so I want to see someone outside Google reproduce it.
Our team used Gemini 4 Argon to build and edit this entire video, then displayed it on a real vintage CRT TV 📺
We used Argon in @Antigravity to script the 3D Blender boot animations and retro SuperCollider audio, then piped the signal via Raspberry Pi to a real TV screen.
SOFIA retweeted
Sam Altman just told an oil and gas conference in Abu Dhabi that OpenAI's next model proved a Millennium Prize problem
His version: 10,000 agents of "the next GPT" worked for about 4 days nonstop. No hierarchy, no bosses. Each agent picked its own direction and shared progress with the rest
There are 7 Millennium Prize problems, $1M each. In 26 years humans have solved one
Two more lines from the same talk:
> "At the start of this year ... AI was mostly not as smart as our smartest scientist and now it's mostly smarter than them"
> On keeping people in control of AI: "I can't believe that this differentiates us from other AI companies but it does"
That model isn't out yet. He says it's "going to be released soon"