Marre

25 | building apps with AI and travel the world by my own ✈️🌎

[Marre @himarre]

@softfusee Consistency is the moat here and it just got cheap. KYC teams should be nervous

[Marre @himarre]

You can now livestream as a COMPLETELY DIFFERENT PERSON in real time, zero artifacts.

A live video feed gets transformed frame by frame to match a reference character's appearance, while your actual motion and face drive it. A small inset shows the raw source feed so you can see the swap happening live.

No render queue. No post editing. The identity stays CONSISTENT the whole time, no flickering face, no melting between frames.

This breaks the last thing that felt safe about live video. If the person streaming can be anyone, on demand, what does a live feed actually prove anymore.

Save this before the demos get scary.

[Marre @himarre]

@shipper_now I ran a landing page through a delve/seamless/robust filter last month and rewrites bumped my email signups by roughly 18% btw

[Marre @himarre]

https://t.co/tl19Odqlwy

[Marre @himarre]

I KILLED MY $200/DAY CLAUDE OPUS BILL AND SWAPPED IN A $20 SUBSCRIPTION MODEL

GPT-5.6 Terra running inside my Hermes agent now beats Claude Opus 4.8 on the work that actually pays: computer use, agentic loops, reliability.

The switch took under a minute.

Open the Hermes dashboard in your terminal. Go to Models. Set MAIN MODEL to GPT-5.6-Terra (or Luna if you're on a lower plan). Then drop one command in Telegram: /reasoning medium.

Not high. Not ultra. MEDIUM.

Everyone burns tokens cranking reasoning to max. On Sol, medium is the sweet spot. Fast, sharp, less overthinking, and it just works instead of stalling on the same task.

The cost math is what broke my brain. I was bleeding $200 a day in API spend. Now it's a flat $20 month plan doing more. My usage graph shows Claude falling off a cliff.

And it doesn't shatter. Every time OpenClaw shipped an update, my agents broke. This hasn't broken once on a model update.

Then I let it run wild:

Built a local AI lab, pulling models off Hugging Face matched to my RTX 5090.

Handed it full autonomous Unity game dev. It shipped.

Set a daily cron job to watch Micro Center GPU prices and ping me.

Same agent. Three jobs. One cheap sub.

The expensive-API era of agents just ended and almost nobody noticed. Bookmark this before your competitor reads it.

[Marre @himarre]

GOOGLE DEEPMIND JUST TURNED A RANDOM OFFICE WALKTHROUGH INTO A SEARCHABLE 4D WORLD YOU CAN QUERY IN PLAIN ENGLISH.

You feed it raw video. Nothing tagged, nothing labeled, no manual annotation.

GenCeption tears that footage apart into four layers at once: a depth map, surface normals, a camera raymap, and segmentation masks. Then it rebuilds the whole space as a 4D scene you can actually talk to.

The demo is stupid simple and that's the point. Type "I want to find a comfortable place to sit" and the sofa lights up. "I need a whiteboard." Whiteboard, found. "Is there a monitor I can use?" Highlighted. "Where's my colleague?" It grounds the person.

No bounding boxes drawn by hand. No labels typed in advance. The model figured out what a sofa IS and where it sits in 3D space, from pixels alone.

Every video you've ever recorded is now a database you can search with a sentence. Robots, AR, security, indexing. Same primitive.

The camera roll just became queryable.

[Marre @himarre]

I KILLED MY $200/DAY CLAUDE OPUS BILL AND SWAPPED IN A $20 SUBSCRIPTION MODEL

GPT-5.6 Terra running inside my Hermes agent now beats Claude Opus 4.8 on the work that actually pays: computer use, agentic loops, reliability.

The switch took under a minute.

Open the Hermes dashboard in your terminal. Go to Models. Set MAIN MODEL to GPT-5.6-Terra (or Luna if you're on a lower plan). Then drop one command in Telegram: /reasoning medium.

Not high. Not ultra. MEDIUM.

Everyone burns tokens cranking reasoning to max. On Sol, medium is the sweet spot. Fast, sharp, less overthinking, and it just works instead of stalling on the same task.

The cost math is what broke my brain. I was bleeding $200 a day in API spend. Now it's a flat $20 month plan doing more. My usage graph shows Claude falling off a cliff.

And it doesn't shatter. Every time OpenClaw shipped an update, my agents broke. This hasn't broken once on a model update.

Then I let it run wild:

Built a local AI lab, pulling models off Hugging Face matched to my RTX 5090.

Handed it full autonomous Unity game dev. It shipped.

Set a daily cron job to watch Micro Center GPU prices and ping me.

Same agent. Three jobs. One cheap sub.

The expensive-API era of agents just ended and almost nobody noticed. Bookmark this before your competitor reads it.

[Marre @himarre]

@rileybrown Ran a few coding tasks through it this morning. The agentic tool use is the real jump, it chains 4-5 steps without losing the thread where K2 would drift by step 3. Context handling feels noticeably tighter too.

[Marre @himarre]

@angeldot_ hahahahahaha finally

[Marre @himarre]

@benyuls so is that means i can finally cancel my claude sub?

[Marre @himarre]

@0x_fokki 95 what??????

[Marre @himarre]

@softfusee this account deserves blue checkmark

[Marre @himarre]

@zaynmcps bookmark this, re-read it before you launch anything

[Marre @himarre]

everyone’s optimizing the product. the ones winning are optimizing who sees it first.

distribution isn’t a growth channel, it’s the actual business model now.

[AffiliateNetwork.com @affiliatenw]

Distribution is the only moat.

[Marre @himarre]

if you’re still burning tokens on every prompt with Claude or GPT you haven’t tried the new open models yet

GLM 5.2 and Kimi K2.7 are hitting frontier benchmarks at a fraction of the cost

Coworker’s new Artifacts mode routes each task to the right model automatically

decks, dashboards, apps, docs, frontier quality at 5x less token spend

https://t.co/thVLfnrXob

[Coworker.ai @coworkerapp]

Introducing Open Artifacts by Coworker.

Build beautiful work products, routed to the right model for every task.

Frontier outputs. Exactly the right context. 80% lower cost.

[Marre @himarre]

@brenzhills i onboarded during covid fully remote. would have killed for this instead of listening to recorded calls for 2 weeks

[Cresta @cresta]

Live customer conversations are no place for agents to learn on the job.

Cresta Training Simulator replaces scripted role-play with live, dynamic AI simulations that help human agents practice before going live.

Learn more ➡️ https://t.co/1AKsZBRfyD

[Marre @himarre]

@ftcarpe la conversión a word siempre rompe el formato original.

[Marre @himarre]

@Giftyjf1 Damn

[Marre @himarre]

@anneyuls Es una pasada, estoy probándolo para mis clientes y va que da gusto

[Marre @himarre]

yo buscando referencias virales para la marca durante 4 horas

mientras tanto, Lightreel ya lo hizo en 10 segundos

[Lightreel AI @lightreelai]

Today, we're introducing GPT 5.6 for marketing

Just enter your brand and Lightreel watches millions of live TikToks and Reels to:

→ find creators
→ write viral scripts
→ see why your videos are flopping

Stop spending hundreds of hours doomscrolling.