Kitsui retweeted
Apple would never show you this: Fable 5.2 just used my free Claude skill to open up 40 years of Macs in one 30-second shot.
i handed my skill, teardown, a ChatGPT render of every Mac and asked for one take from 1984 to 2024 with no cuts.
what my skill does with them:
> the renders only give the look, Claude writes every case, logic board and fan as three.js code
> nothing moves until its own still sits next to the picture and the two line up
> one timeline drives the whole shot: the Macs, the fan count, the weight bars and the sound all tick on the same clock
> it never watches the video, it grades a grid of frames at every beat and patches whatever is off
> the jump from one Mac to the next is code too, a blue sweep that rebuilds the case
> my GPU captures the take at 60 fps and a python script scores it on the same beats
i built this skill and i still didn't know the G5 needed 9 fans until it opened one up.
the skill and every prompt are in my article below 👇
I just dragged Blender to the trash. I built a free Claude skill and Fable 5.2 just used it to take a robot octopus apart joint by joint in one 24-second shot.
i gave my own skill a single ChatGPT render and asked it to open every joint on camera.
what came apart:
> Claude took only the look from the render and coded every plate, roller and gear in three.js
> the 2 camera eyes drop their sapphire caps and open a 10-leaf aperture inside a ring of 16 IR LEDs
> the dome folds back in 8 hinged plates to show a 7-cylinder pump on a 44:30 gear pair
> each of the 8 arms runs a 3-axis shoulder, 2 rows of grip rollers and a 16-segment tip that coils and swivels 32° either way
> the thruster pulls apart into 3 nested sleeves around a 12-vane rotor and underneath a two-part jaw opens on an 18-tooth grinder
> on full teardown every part floats loose through a 330° turn, then the body closes up, the thruster lights and it heads out of frame
i know every line this skill runs and i still keep replaying the teardown.
the skill and the exact prompts i used 👇
I just dragged Blender to the trash. I built a free Claude skill and Fable 5.2 just used it to take a robot octopus apart joint by joint in one 24-second shot.
i gave my own skill a single ChatGPT render and asked it to open every joint on camera.
what came apart:
> Claude took only the look from the render and coded every plate, roller and gear in three.js
> the 2 camera eyes drop their sapphire caps and open a 10-leaf aperture inside a ring of 16 IR LEDs
> the dome folds back in 8 hinged plates to show a 7-cylinder pump on a 44:30 gear pair
> each of the 8 arms runs a 3-axis shoulder, 2 rows of grip rollers and a 16-segment tip that coils and swivels 32° either way
> the thruster pulls apart into 3 nested sleeves around a 12-vane rotor and underneath a two-part jaw opens on an 18-tooth grinder
> on full teardown every part floats loose through a 330° turn, then the body closes up, the thruster lights and it heads out of frame
i know every line this skill runs and i still keep replaying the teardown.
the skill and the exact prompts i used 👇
Fable 5.2 has no business being this good. It built a 3.4 km sand monster from ONE picture and sent it 200 m under the dunes.
I gave it the picture and 28 seconds later the thing had surfaced, shown off all 72 legs and dived back under.
It all runs on teardown, a free open-source Claude Code skill another dev put out (nearly 3,500+ stars on GitHub). Claude works out how the creature moves before it draws a single plate.
Why rig 72 legs by hand when Claude writes the whole gait as code?
Here's how it plays out: from the picture Claude writes the leg wave as math first, so the feet-down counter, the gait chart and the depth profile all read the same clock as the body. It checks a still at every beat, fixes what's off and records at 60 fps: the boring crown breaches with 78% of the body still under sand, 36 leg pairs ripple down a 3,420 m chain, x-ray lights up a core chain of 7 nodes and the head dives back to -218 m with 92% of the body buried.
Step-by-step setup:
1. Grab teardown with one npx line
2. Open Claude Code and type /teardown
3. Give it one picture of the creature
4. Let it sign off each still before filming
5. Hit render and watch it crawl
It all stays in Claude Code and nobody animates a single leg by hand.
The setup and every prompt are waiting in the article below.
Kitsui retweeted
f*cking gold. my AI employee read my GitHub top to bottom and came back with 6 issues people had opened on my repo that i never once answered
i only asked him to take my weekly routine off my hands
this ai works out of my Slack. GitHub is plugged in read-only, so he sees everything and can't change a line
first thing he did was write my rule into the channel so it sticks:
"don't post, commit or send anything without my yes"
then he worked through the week:
→ news sweep: 5 real releases, 3 dropped with a reason for each
→ hooks: the same request on 3 brains, with the price of every run
→ repo: 6 open issues, 14 outside links checked, 0 changes made
→ release table: he found his own wrong date and fixed it
→ monday run: rebuilt as an automation, 5 of its 8 steps plain scripts
→ end of day: a credit-by-credit breakdown of what all of it cost
then the line i didn't expect from software. his test run cost 196 credits after he'd estimated under 58, and he said it before i could: "I was wrong earlier."
13 messages. 12 jobs. 6 forgotten issues. 0 things done behind my back
watch around second 13, the estimate gets corrected right on screen
how many issues are sitting unanswered in yours?
my rule, the setup and every message are in the article ↓
me when Breken finds bug number 3 and my 200% refund is gone
Introducing Breken: Autonomous Production Engineering
Software creation is a solved problem. Verifying it works takes months.
Recreating the complexity of production remains one of engineering’s hardest problems. We’ve solved it.
Breken runs your actual system in a virtual world of synthetic users, agents and failing services, running millions of scenarios to find what breaks before your customers do.
breken.ai
me when Breken finds bug number 3 and my 200% refund is gone
Introducing Breken: Autonomous Production Engineering
Software creation is a solved problem. Verifying it works takes months.
Recreating the complexity of production remains one of engineering’s hardest problems. We’ve solved it.
Breken runs your actual system in a virtual world of synthetic users, agents and failing services, running millions of scenarios to find what breaks before your customers do.
breken.ai
I'm actually scared now: Fable 5.2 took a $6,950 NVIDIA AI box apart layer by layer from ONE picture.
All I wanted was to see how a desktop supercomputer fits together, so I ran the free skill from the article below on a single image.
Here's what it pulled out:
> Every rotor, pipe and cable is three.js code, built off that one picture.
> Twin 13-blade rotors and a 74-slot exhaust keep 1 petaflop of FP4 compute running on a 240 W supply.
> Lime fluid runs from a glass tank and twin pumps through every pipe, then 21 fins carry the heat out.
> Under it all sits one package: 20 Arm cores and a Blackwell GPU, enough to hold a 200B-parameter model.
A dev put this whole method out as a free Claude skill. Give it a month and everyone will be posting these.
Grab the skill and every prompt in the article below👇
A 20-year-old street photographer from Chengdu spent $10.50 and one Sunday shooting 50 strangers against a white wall. None of them knew what it was for.
By morning he had a 30-second clip of Shibuya crossing with 2 million views on it.
The trick was simple. He shot the people himself and generated the street around them. 50 strangers from a wall in Chengdu now cross Shibuya at rush hour.
He didn't chase a random idea.
He took a structure that was already spreading: a place everybody knows, an ordinary minute in it, one thing in frame that should not be walking around.
Most AI clips hold 2 or 3 references. This one holds 50 and no face drifts between shots.
The $10.50 went to Picsart with Seedance 2.5 inside
↳ picsart.go.link/a2S32
He posted it on TikTok, Reels and Shorts the same night.
Then he kept the crossing and swapped the crowd. New faces, new costumes, same 30 seconds.
One Sunday of shooting. 50 files that keep going back to work.
The passport template is in the article 👇
🚨 STOP PAYING CLAUDE OPUS 5.5 TO READ INVOICES. YOU'RE OVERPAYING ×261.
1M invoices cost $4,368 on Opus 5.5 at list prices. XERJ Reader does the same pile for $16.75.
It doesn't type. One pass and it pulls the exact words out of the document, start and end attached. Output tokens cost $0.
> 198 ms for one answer, network included
> Opus 5.5 needs 5 s just for its first token
> 219 ms for 6 fields off an invoice, one call
> ask what the invoice never says and you get "No answer" instead of a guess
> GPT-6 Astra bills $10,920 for the same 1M
The catch is long files. A 420,721-character one took 26 s and it still dug the answer out at character 415,880.
The waitlist opened today.
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.
Fable 5.2 got scary good at taking things apart. Almost nobody noticed.
I gave it one picture of an instant camera and got back one that shoots, ejects a print and comes apart in 24 seconds.
The trick is teardown, a free open-source Claude Code skill from another dev (3,100+ stars on GitHub). It makes Claude figure out how the thing works before it builds a single part.
Why rig a camera in Blender for a week when Claude can write every gear as code?
Here's the flow: from that one picture Claude writes the mechanism as math first, so the shot counter, the flash charge and the developing bars track the same clock as the model. It checks a still at every beat, fixes what's off and records the tour at 60 fps: a fresh pack of 8 slides in, the shutter fires at 1/125 s shown slowed, a 45° mirror folds the light down and 2 steel rollers 1 mm apart burst the paste pod as the print comes out.
Step-by-step setup:
1. Add teardown with one npx line
2. Open Claude Code and type /teardown
3. Hand it a picture of the gadget you're curious about
4. Let it check each still before filming starts
5. Hit record and watch every part come out
It all lives inside Claude Code. The keyframing part of the job is gone.
Full setup and prompts are in the article below.
Fable 5.2 got scary good at taking things apart. Almost nobody noticed.
I gave it one picture of an instant camera and got back one that shoots, ejects a print and comes apart in 24 seconds.
The trick is teardown, a free open-source Claude Code skill from another dev (3,100+ stars on GitHub). It makes Claude figure out how the thing works before it builds a single part.
Why rig a camera in Blender for a week when Claude can write every gear as code?
Here's the flow: from that one picture Claude writes the mechanism as math first, so the shot counter, the flash charge and the developing bars track the same clock as the model. It checks a still at every beat, fixes what's off and records the tour at 60 fps: a fresh pack of 8 slides in, the shutter fires at 1/125 s shown slowed, a 45° mirror folds the light down and 2 steel rollers 1 mm apart burst the paste pod as the print comes out.
Step-by-step setup:
1. Add teardown with one npx line
2. Open Claude Code and type /teardown
3. Hand it a picture of the gadget you're curious about
4. Let it check each still before filming starts
5. Hit record and watch every part come out
It all lives inside Claude Code. The keyframing part of the job is gone.
Full setup and prompts are in the article below.
Kitsui retweeted
Love that! I wish my receipts had always sorted themselves like this.
I just tested Univer SDK on a boring job: turning a pile of receipts into an expense sheet.
before this, AI could read receipts but it turned smudged 3s into 8s and counted the same receipt twice, and fixing that meant copying everything into Excel and redoing the totals by hand.
with Univer SDK the expense sheet lives inside the app, the AI writes into real cells and every row checks itself with a formula.
[the whole run is in the 60 seconds above, here is the short version]
1. 24 receipts go in, the agent writes each one into the sheet through Univer's API
2. a check formula compares subtotal + tax with the total and flags row 19: $84.10 against a $34.10 receipt
3. I type 34.10 into that cell, the check turns OK and the total drops from $1,128.39 to $1,078.39
4. the total formula skips the receipt uploaded twice, so the duplicate never gets counted
5. I move one coworking day pass from Office to Travel and $54.44 shifts between categories on the spot
the agent and I work in one sheet, and Univer recalculates every total after each edit.
Univer SDK is open source and drops this spreadsheet into any web app.
the live demo is in the reply below.
Univer on GitHub: github.com/dream-num/univer
Can't stop watching this! Never thought a robot dog could be this fun to take apart.
All I did was ask Claude Opus 5.5 for a fun teardown of a robot dog, using the skill from the article below.
What came back:
> It built the whole dog from one picture. Every motor, camera and panel is three.js code.
> The head slides off on 5 pairs of stereo cameras: that's how it sees 360° around itself.
> Each leg runs 3 motors, 2 at the hip and 1 at the knee, 12 in all.
> Pull the battery out of the belly and the step rate drops to 0 until it's back in.
A dev packed the whole process into a free Claude skill, so anyone can build their own from one picture.
The skill and the prompts are in the article below👇
The amount of time I've spent staring at a blank screen trying to make one decent ad... I'd rather not talk about it.
Dreamina Marketing Studio is now live!✨
Meet Dreamina Marketing Studio, your all-in-one AI workspace for creating professional marketing content. Bring your creative ideas to life, produce marketing assets, and adapt content for different channels — all in one place.
✨ What can you create with Marketing Studio?
🎬 Video Recreator
▸ Use your favorite videos as inspiration. Add your own products and characters to create ad videos that are uniquely yours in no time.
👤 AI Avatar Creation
▸ Choose from a rich library of AI avatars, or upload or generate your own to create engaging product showcases and marketing content.
🔄 Swap Products Without Reshoots
▸ Replace products or characters without starting from scratch. Use Motion Transfer to preserve the original movement, or Object Swap to seamlessly replace subjects in a scene.
🎨 One-Click Marketing Asset Creation
▸ Choose industry-specific presets and turn simple prompts into polished product photos, promotional posters, images, and videos — no professional design skills needed.
🌍 Multi-Channel Content Adaptation
▸ Adapt aspect ratios and compositions for different platforms. Translate text in images and dub videos into multiple languages to reach more audiences.
📈 Scale Ad Creation with AI
▸ Let the Marketing Agent generate scripts, create ad videos in batches, and test different variations to identify top-performing creatives.
🔥 Limited-Time Offers Across Dreamina
▸ Dreamina Basic: just $0.29 for your first month.
▸ Advanced or Ultra: 50% off your first month on a monthly plan, or your first year on an annual plan.
▸ Dreamina Seedance 2.5: up to 50% fewer credits per generation.
🎁 12-Hour Giveaway: Repost & Comment to Get a Bonus!
▸ The first 50 businesses to follow us, repost this post, and leave a comment using their official accounts will each get one free month of standard membership. First come, first served!
#DreaminaAI #DreaminaMarketingStudio
Love this! I wish I'd had it back when I was animating 3D characters by hand.
I asked Opus 5.5 to bring a golem from Tripo's gallery to life. It called Tripo CLI, gave the golem a skeleton with 78 bones, then built the ruins, the fog and the moonlight around it.
Opus runs the job. Tripo makes it move.
Try Tripo with code SCCCSA, details in the first reply.
holy f*ck, THIS IS WHY 3D PRODUCT STUDIOS SHOULD BE NERVOUS RIGHT NOW.
what you're seeing is a Vision Pro Claude rebuilt from a single image, pulling itself apart piece after piece.
one timeline file decides when each part comes out and what every panel shows.
the best part?
anyone can make one with Claude Opus 5.5 + three.js + a free skill.
the recipe:
1. CHATGPT - GENERATE THE REFERENCE
one image fixes the style: chrome frame, dark glass, orange trim and a knit band.
2. CLAUDE OPUS 5.5 - CODE THE HEADSET
glass, lenses, 12 cameras and the chip written in three.js and compared to the picture before any motion.
3. ONE TIMELINE - DIRECT EVERY BEAT
glass, lenses, cameras, chip, fan, band: each piece gets its moment and the panels follow the same clock.
4. YOUR GPU - RENDER FRAME BY FRAME
60 fps, so every tiny camera module stays crisp as it drops out.
what comes out?
24 seconds where the camera card flips from ACTIVE to DETACHED the moment the shell splits.
step by step, it's all in the article 👇
holy f*ck, THIS IS WHY 3D PRODUCT STUDIOS SHOULD BE NERVOUS RIGHT NOW.
what you're seeing is a Vision Pro Claude rebuilt from a single image, pulling itself apart piece after piece.
one timeline file decides when each part comes out and what every panel shows.
the best part?
anyone can make one with Claude Opus 5.5 + three.js + a free skill.
the recipe:
1. CHATGPT - GENERATE THE REFERENCE
one image fixes the style: chrome frame, dark glass, orange trim and a knit band.
2. CLAUDE OPUS 5.5 - CODE THE HEADSET
glass, lenses, 12 cameras and the chip written in three.js and compared to the picture before any motion.
3. ONE TIMELINE - DIRECT EVERY BEAT
glass, lenses, cameras, chip, fan, band: each piece gets its moment and the panels follow the same clock.
4. YOUR GPU - RENDER FRAME BY FRAME
60 fps, so every tiny camera module stays crisp as it drops out.
what comes out?
24 seconds where the camera card flips from ACTIVE to DETACHED the moment the shell splits.
step by step, it's all in the article 👇
I'M F*CKING SPEECHLESS. A free Claude skill turned one picture into a 1,300 m sand worm that dives past 600 m under the dunes and comes back up.
i give it a week before somebody puts a paywall in front of this.
under the sand:
> the whole worm is three.js: segmented rings, each with an armor plate, a hydraulic actuator, a drive piston and a sensor node
> the head spins 36 displacement blades on a drill at 10.8 RPM behind a hammer ring pounding at 15.3 Hz
> it crawls on a peristaltic ring drive, the body bunches and stretches while the locomotion chart steps through 4 phases
> every time it breaches the hammer goes idle and the blades stow, then they spin back up as it goes under
> hit X-RAY and the shell turns to glass while the depth readout runs down to -610 m
> 7 camera views sit on one bar from cine to x-ray while the surface map tracks it through the whole loop
next i'm feeding it an oil rig and posting whatever crawls out.
the full walkthrough and the skill, below 👇
I'M F*CKING SPEECHLESS. A free Claude skill turned one picture into a 1,300 m sand worm that dives past 600 m under the dunes and comes back up.
i give it a week before somebody puts a paywall in front of this.
under the sand:
> the whole worm is three.js: segmented rings, each with an armor plate, a hydraulic actuator, a drive piston and a sensor node
> the head spins 36 displacement blades on a drill at 10.8 RPM behind a hammer ring pounding at 15.3 Hz
> it crawls on a peristaltic ring drive, the body bunches and stretches while the locomotion chart steps through 4 phases
> every time it breaches the hammer goes idle and the blades stow, then they spin back up as it goes under
> hit X-RAY and the shell turns to glass while the depth readout runs down to -610 m
> 7 camera views sit on one bar from cine to x-ray while the surface map tracks it through the whole loop
next i'm feeding it an oil rig and posting whatever crawls out.
the full walkthrough and the skill, below 👇














