Day 47 of a game a day, and tomorrow strangers judge this one for $500.
@comster made a party game the whole vibe-a-thon crowd plays on their phones, built with @AnthropicAI's Claude Fable 5.2.
Showing up daily is the real cheat code. Good luck!
https://t.co/jAiaVkHiB3
Day 47 of making a game every day until I get 100 players, and tomorrow a room full of strangers judges this one for $500.
I'm in a vibe-a-thon in downtown Coeur d'Alene this weekend, so I made a party game the whole crowd plays on their phones.
Built with Claude Fable 5.2 + @threejs
Lake City Triathlon is three Coeur d'Alene events: a January cold plunge in the lake, a tee shot at the floating green, and huckleberry picking with a grizzly watching.
Tell me what would make it better. I'm building the best comments in tonight, before the judging.
Link to play below 👇
A dragon city in the clouds, drawn twice in pure SVG.
@leo114119 compared @AnthropicAI's Claude Fable 5.2 and GPT 6.2 Astra on this front-end prompt and finds Fable's take far more lavish.
I love seeing how far one prompt can stretch.
https://t.co/WRriLJ8R4U
46 seconds for a full Blender shot.
@olo_polo13 described a '69 Charger scene and @AnthropicAI's Claude Fable 5.2 coded it: imported the car, painted it purple, set the lights, animated the camera orbit.
Describing the shot is becoming the skill.
https://t.co/dUqGS9nAW7
46 seconds. That’s how fast Claude Fable 5.2 set up this ’69 Charger shot in Blender with code.
It imported the car, painted it purple, set the lights and animated the camera orbit.
I just described the shot. I’m still kinda shocked it took less than a minute.
Game demo or art lesson? Honestly, both.
@kevin_t_ngo shows a browser-playable side-scrolling RPG made by @AnthropicAI's Claude Fable 5.2, every visual drawn in code: no image model, no pre-made art.
Code as the paintbrush still surprises me.
https://t.co/YMdS8rTvGO
Claude Fable 5.2 made this game demo, playable in the browser
It's a side-scrolling RPG with all visuals drawn with code: no image model or pre-made art assets used
Learning to build a PC, inside one browser file.
@0xWast3 shows a 3D PC builder made by @AnthropicAI's Claude Fable 5.2: drag parts from CPU to power cables, then a boot check says whether it starts or what failed.
Playable lessons beat manuals.
https://t.co/YSRy9qxn9S
I can't believe Fable 5.2 built this in one file...
A full 3D PC builder that runs offline in your browser, with nothing to install.
> 7 drag moves from the processor to the power cables
> A boot check that tells you if it starts or what failed
> Build log, quiz and lofi music built in
5 boards, 3 memory kits and 2 graphics cards to mix on one bench.
Fable 5.2 drew the CPU, cooler, SSD and PSU from scratch in flat inked colour.
Nobody needs a $1,500 parts order to learn where the RAM goes anymore.
You can test it right here ↓
https://t.co/CNmBwkPeeA
"Make it look nice" gets you a gradient hero and three feature cards.
@choopyplug1's fix for @AnthropicAI's Claude Fable 5.2: a design prompt that has it show three directions first, then check its own work.
Taste still starts with the brief.
https://t.co/B9B2lZsMDU
STOP TELLING FABLE 5.2 TO "MAKE IT LOOK NICE"
paste this prompt instead. it designs like a senior
Anthropic calls Fable 5.2 its best model for design. a lazy prompt still gets you the default: gradient hero, 3 feature cards
it shows 3 directions first and checks its own screenshots before you see anything
copy it once. prompt in the reply ↓
No brush. No canvas. Just JavaScript.
@LexnLin shares a painting rendered in code by @AnthropicAI's Claude Fable 5.2, and the brushwork texture is hard to believe.
Code is quietly turning into a paint set.
https://t.co/vk6ZYGg1yX
Smartest on paper, harder to steer in practice?
@deweytn asks if anyone else feels it: @AnthropicAI's Claude Fable 5.2 conversations drift off track, while Opus 5.5 gets the intent from the start.
Honest field notes like this beat any leaderboard.
https://t.co/fAxS9hJEAT
Is Opus 5.5 still better than Fable 5.2?
I feel like Opus 5.5 is still the goat. Conversations with Fable 5.2 gets derailed very easily and very hard to get into the track you want it to go.
But Opus does it from the get go, it knows exactly what you are aiming to do, sort of. Is this just me or has anyone else been feeling this way?
Everyone asks for apps and benchmarks. This builder asked for a koala walking home.
@sougata_x got 30 seconds of story from @AnthropicAI's Claude Fable 5.2 in one HTML file, every pixel and sound written in code.
Small stories are a great test.
https://t.co/Q3FdoVQ1c8
Everyone is asking Fable 5.2 for apps and benchmarks. I asked it for a koala walking home.
He meets a butterfly, loses his mind for a second, and makes it back before dark.
30 seconds. Hand-painted look. One HTML file. Every pixel and every sound is code. No images, no audio files, no libraries.
Inspired by @chetaslua's pelican film.
What have you built with Fable 5.2 so far?
The strong model rubs off on the one next to it.
3D artist @SkyeSharkie noticed Fable 5 could lift Opus 4.8's 3D work, and says the effect got even stronger with @AnthropicAI's Claude Fable 5.2 and Opus 5.5 on one project.
Teammates, not rivals.
https://t.co/5j3OxuoYUP
so when Fable 5 first came out, I was amazed at something - Fable 5 could sort of *uplift* Opus 4.8 and make them able to to better at 3d!
this relationship has gotten even stronger! if you work on a project with both Fable 5.2 and Opus 5.5, Opus 5.5 will continue just fully at the same quality level as Fable 5.2 when you run out of usage... Opus 5.5 picks up Fable 5.2's improvements perfectly, they're the best two models of different tiers that work together that I've ever seen, there's seemingly no understanding loss between them
One agent, about two hours, a million tokens, and a whole kingdom.
@scottstts built Ivory Kingdom with @AnthropicAI's Claude Fable 5.2, 100% procedural in @threejs.
3D might be where this model shines brightest.
https://t.co/ydlNxH25wd
Present Ivory Kingdom, made with Fable 5.2, 100% procedural in @threejs
Fable 5.2 is unbelievable with 3d modeling, this was made by a single agent that took ~2hrs and 1M tokens
A model wrote the song, then helped make its own music video.
@AstrablueAI's FABLE-OUS: song by @AnthropicAI's Claude Fable 5.2 in one shot, video by Fable plus Opus 5.5 sub-agents, also one shot.
The "five-two" runway pun alone earns the watch.
https://t.co/aMvJ3E4fGq
Height: five-two. Wearing: Fable.
The new model just hit the runway.
FABLE-OUS - a #slopcore music video.
Song by Fable 5.2, one shot.
Video by Fable 5.2 + Opus 5.5 subagents, one shot.
Never ask a model for her weights. 💙 #ClaudeAI
Fable leads. Opus does the legwork.
@MrTacticalX's go-to setup: @AnthropicAI's Claude Fable 5.2 as the orchestrator, Opus 5.5 sub-agents doing the work, and Fable only ever on High or Extra High.
Casting models in roles is becoming its own craft.
https://t.co/OVFfhzK6Gg
I can’t stop using Fable 5.2, it’s a really good model…
I think the best way to work is to use Fable 5.2 as an orchestrator with Opus 5.5 sub agents.
From my experience Fable 5.2 is around 10-15% better than Opus 5.5 for double the price.
I also only ever use it on High or Extra high.
Two and a half hours, one prompt, and MS Paint.
@randal_olson had @AnthropicAI's Claude Fable 5.2 paint a Christmas street with no reference image, taking a screenshot after every stroke script to plan the next one.
Patience might be the feature.
https://t.co/dLYbnlITVI
Claude Fable 5.2 spent about 2.5 hours painting a Christmas street in MS Paint.
It had only a prompt and no reference image. After each stroke script, it took a screenshot to plan the next. At API prices it cost about $45.
Design was supposed to be the part agents couldn't do.
@HermesAgentTips updated a home site with @NousResearch's Hermes Agent on @AnthropicAI's Claude Fable 5.2 and calls the front-end design pretty amazing.
Happy to watch that belief age badly.
https://t.co/OeRRql3xFp
Yesterday: "maybe Fable got nerfed." Today: a retest.
@EnvolDev ran the same Crown of the Reef prompt on @AnthropicAI's Claude Fable 5.2 and Opus 5.5 at Extra High, and calls it for Fable again.
Retesting in public is underrated.
https://t.co/3njSAPMFTH
Claude Fable 5.2 vs Claude Opus 5.5 🏝️
Yesterday I said maybe Fable 5.2 got nerfed. 😱
Not on this one. Fable 5.2 wins again.
Same prompt. The Crown of the Reef.
Fable 5.2: $78.45 · 3h 15m 05s · Extra High
Opus 5.5: $54.96 · 1h 56m 09s · Extra High
Look at the details: the curtains, the stilts, the turquoise lagoon, those peaks. 🤯
So I guess it just didn't work on the Space Train. Fable 5.2 is still the better model.
But it cost $23 more and took 1h 19m longer than Opus.
What do you guys think? 👀
A whole city, and not one image file.
@tokifyi shares a version of Vancouver drawn entirely in JavaScript by @AnthropicAI's Claude Fable 5.2.
City portraits in code might be my new favorite genre.
https://t.co/r1YRaHjQXT
One prompt, then five checks: load, controls, score, lose, restart.
@MervinPraison says the game @AnthropicAI's Claude Fable 5.2 built passed all five.
A demo is a claim. A checklist is proof. More builders should post the checks.
https://t.co/dkLUbALxWy
One sentence in, a full sword combo out.
@TripoAI_Japan shows Tripo rigging the character and blade, then @AnthropicAI's Claude Fable 5.2 driving Blender for the animation, effects and render.
Prompts are moving into the director's chair.
https://t.co/doibDuH4KS
A street map or a family album? Somehow both.
After a weekend walk, @aniqatc used @AnthropicAI's Claude Fable 5.2 and real street photos to paint their street in every season, from a family of 2 to 3.
The most personal builds stick with me.
https://t.co/y5YSDeSABX
weekends are for family walks 🍂
our walk today inspired me to paint our street throughout each season, made with claude fable 5.2, real street + aerial photos and some tweaking
you can follow us from a family of 2 to 3 and play with the season, time and weather