Pinned post
Latest posts 5
this is just crazy...
The model reading every licence in a 3D asset library never writes a word of its own, and that is why it can be trusted.
Asset licences bury the rule that matters in plain prose, so a search for commercial walks right past it. Same with invoices, contracts, claim emails, shipping notifications, logs your bot burns tokens just pulling a single field out.
5 jobs you’re wasting tokens on right now:
1. pulling the total off an invoice
2. finding the notice period in a contract
3. grabbing the policy number from a claim email
4. naming the failing order in a log
5. reading a date out of a shipping email
XERJ Reader does all 5 and every asset licence in your library, for $0.01 per 1M input tokens, output free.
> ASK – one question in your own words
> READ – PDF, Word and HTML, no schema
> SCAN – every licence in one pass
> MARK – the exact clause and its position
> NONE – a plain no when nothing answers
Ask whether an asset can ship in a paid game, and XERJ Reader highlights the clause in the source file itself. The answer comes back word for word with its exact spot in the doc. It never paraphrases, the words are already sitting in the document, so there is nothing to invent.
That is around 1,050x below Claude Opus 5.5 and 26x below Haiku 5.5, with each answer back in about 80 ms in batch.
Cheap is not the point with assets, where one made up permission turns into a takedown after a game ships. Hand the extraction off and keep your LLM for the thinking.
The waitlist opens today, and the launch video is below.
The model reading every licence in a 3D asset library never writes a word of its own, and that is why it can be trusted.
Asset licences bury the rule that matters in plain prose, so a search for commercial walks right past it. Same with invoices, contracts, claim emails, shipping notifications, logs your bot burns tokens just pulling a single field out.
5 jobs you’re wasting tokens on right now:
1. pulling the total off an invoice
2. finding the notice period in a contract
3. grabbing the policy number from a claim email
4. naming the failing order in a log
5. reading a date out of a shipping email
XERJ Reader does all 5 and every asset licence in your library, for $0.01 per 1M input tokens, output free.
> ASK – one question in your own words
> READ – PDF, Word and HTML, no schema
> SCAN – every licence in one pass
> MARK – the exact clause and its position
> NONE – a plain no when nothing answers
Ask whether an asset can ship in a paid game, and XERJ Reader highlights the clause in the source file itself. The answer comes back word for word with its exact spot in the doc. It never paraphrases, the words are already sitting in the document, so there is nothing to invent.
That is around 1,050x below Claude Opus 5.5 and 26x below Haiku 5.5, with each answer back in about 80 ms in batch.
Cheap is not the point with assets, where one made up permission turns into a takedown after a game ships. Hand the extraction off and keep your LLM for the thinking.
The waitlist opens today, and the launch video is below.
0:13
I've built a new AI model that is not allowed to generate text. Instead, it produces a verifiable program that answers your prompt using the original reference.
XERJ Reader-1 is a neurosymbolic model for working with data. It points to the exact words in your document, writes a small program over them, and exact code computes the answer:
• 20-90x faster
• 10-1,000x cheaper (output tokens free)
• every answer is a program you can replay
The shortest path to automating data work.View quoted post on X
I still can't believe what my pc just pulled off... I gave Gemini and Claude a single prompt, and it generated a complete, fully rigged 3D game character with custom animations from scratch.
Not just a static picture. A full turnaround sheet, exported straight into a 3D engine as a "Mushroom King" ready for a Roblox game, complete with walking, running, and special attack states.
It happened in under a minute while I was sitting at my desk testing workflow loops.
My project partner's design concept, text prompts handled by the AI, organic sculpted shapes, low-poly cartoon style, and exact keyframe animations for idle, jump, and combat, all generated seamlessly in one go.
I started clicking through the settings panel. Next to the character model, I saw the full animation tree: Cap Slam, Spore Cloud, Mushroom Roll. Every single action fully parameterized and ready to import.
No tedious 3D modeling for weeks in Blender. No manual rigging or weight painting. Just one prompt, and the AI built a production-ready game asset with active physics and states.
I saved the project files and closed the editor. Who needs a whole design team for rapid prototyping anymore, I don't know. Tomorrow I have to show this to the team.
Any game dev pipeline you're currently running can do the same. If you want to test yours, open up Gemini alongside your design agent, feed it a detailed character concept prompt, and watch it generate complete 3D assets in seconds.
Not just a static picture. A full turnaround sheet, exported straight into a 3D engine as a "Mushroom King" ready for a Roblox game, complete with walking, running, and special attack states.
It happened in under a minute while I was sitting at my desk testing workflow loops.
My project partner's design concept, text prompts handled by the AI, organic sculpted shapes, low-poly cartoon style, and exact keyframe animations for idle, jump, and combat, all generated seamlessly in one go.
I started clicking through the settings panel. Next to the character model, I saw the full animation tree: Cap Slam, Spore Cloud, Mushroom Roll. Every single action fully parameterized and ready to import.
No tedious 3D modeling for weeks in Blender. No manual rigging or weight painting. Just one prompt, and the AI built a production-ready game asset with active physics and states.
I saved the project files and closed the editor. Who needs a whole design team for rapid prototyping anymore, I don't know. Tomorrow I have to show this to the team.
Any game dev pipeline you're currently running can do the same. If you want to test yours, open up Gemini alongside your design agent, feed it a detailed character concept prompt, and watch it generate complete 3D assets in seconds.
0:16
View quoted post on X
HE LEVELED UP THE COCA-COLA TRUCK DISASTER WITH A FULL-SCALE MENTOS LAUNCHER
800k views. 40k likes. 1320 comments. 30 seconds.
filmed from a car window on a highway, "Mentos" on screen. a leaking red coca-cola tanker spills soda across the asphalt while a guy fills a large jug from the stream, pours it over his head, runs to a parked mentos truck, dumps a box of mints into the pool, and triggers a massive chemical geyser that blasts him straight up into the clouds
why it works:
→ fluid physics and high-pressure streams are notoriously brutal to render consistently, but the liquid interaction, heavy splashes, and liquid weight look totally organic as it pours and pools
→ the sequence is physically paced right: heavy leak, gathering liquid, dumping the mints into the puddle, and only then does the massive pressurized eruption trigger. the reaction behaves just like real-world chemistry
→ the camera is static from inside a car with the dashboard cutting into the bottom edge, capturing the chaotic highway incident naturally as an accidental bystander POV
→ overcast daylight with heavy roadside spray creates continuous soft lighting and mist that swallows minor artifacts or generation drift
→ one continuous chain reaction from start to finish: leak to reaction to takeoff. no plot hops, no second location
the takeaway: pick a chaotic crossover event people instantly recognize. they know exactly how the classic Coca-Cola and Mentos reaction goes, so when it delivers on a cinematic scale, they stop checking for flaws
single-event physics simulations like this are what seedance 2.5 does best
800k views. 40k likes. 1320 comments. 30 seconds.
filmed from a car window on a highway, "Mentos" on screen. a leaking red coca-cola tanker spills soda across the asphalt while a guy fills a large jug from the stream, pours it over his head, runs to a parked mentos truck, dumps a box of mints into the pool, and triggers a massive chemical geyser that blasts him straight up into the clouds
why it works:
→ fluid physics and high-pressure streams are notoriously brutal to render consistently, but the liquid interaction, heavy splashes, and liquid weight look totally organic as it pours and pools
→ the sequence is physically paced right: heavy leak, gathering liquid, dumping the mints into the puddle, and only then does the massive pressurized eruption trigger. the reaction behaves just like real-world chemistry
→ the camera is static from inside a car with the dashboard cutting into the bottom edge, capturing the chaotic highway incident naturally as an accidental bystander POV
→ overcast daylight with heavy roadside spray creates continuous soft lighting and mist that swallows minor artifacts or generation drift
→ one continuous chain reaction from start to finish: leak to reaction to takeoff. no plot hops, no second location
the takeaway: pick a chaotic crossover event people instantly recognize. they know exactly how the classic Coca-Cola and Mentos reaction goes, so when it delivers on a cinematic scale, they stop checking for flaws
single-event physics simulations like this are what seedance 2.5 does best
0:30
View quoted post on X
