Maivar

Worlds Builder. AI Art AI Dev .

[Alex @alexfredo87]

If you guys want to understand how I’m doing this, basically I use ImageGen of Astra to create realistic versions of the scene projected onto the geometry from different angles. Then I ask Astra to create a script that isolates and displays the appropriate AI render depending on where the player is looking. It works really well with Three.js.

I hope that makes sense it’s a bit difficult to explain clearly! https://t.co/dfloZn64aT

[Maivar @RadarDiplo]

@RaphaAssouline vraie question : il n'y a qu'a compter les banques détenues par les juifs et le reste pour savoir non ?

[Stefan 3D AI @Stefan_3D_AI]

This is how Opus 5.5 in Claude Code built a WoW-style game in 36h.

There's a lot more to it than one prompt. I wrote a detailed spec, gave it access to Unity, Blender and Higgsfield, and packed my own workflow into skills. I go through all of it in the video: https://t.co/K5VMLms1Ww

[Olivier Rimmel 🕊️ @OlivierRimmel]

La Chine a 3000 ans d'avance!

[Alex Graveley @alexgraveley]

Excited to announce UFO: a new scalable multiplayer agent orchestrator and memory system.

We built from scratch to target pain points we saw with people running agent-first businesses.

Give it a try and let me know what you think :)

[Josh Whiton @joshwhiton]

the surprise conclusion of the machine consciousness debate will be that people are machines.

[Vox @Voxyz_ai]

Everyone's posting motion videos made with Opus 5.5, and they're wild. To make one yourself in Claude Code, give it 𝗮 𝟰-𝗽𝗲𝗿𝘀𝗼𝗻 𝘁𝗲𝗮𝗺, with each subagent owning one part end to end:

→ Breakdown (medium): tile frames from a reference video you love into one image, pull out its pacing, type and transitions, and write them up as HyperFrames code. Find references on https://t.co/6WlRF4Ln7m, which has 2,000+ product launch videos
→ Storyboard (medium): lay out 3 storyboard options in one image for you to pick from, then make one still per scene for the one you choose
→ Build (main session, high): write every scene as HyperFrames code, add music, and render a draft before the final
→ Review (high): watch the render second by second and list the problems by severity, without touching anything

You just 𝗰𝗮𝗹𝗹 𝘁𝗵𝗲 𝘀𝗵𝗼𝘁𝘀 𝗹𝗶𝗸𝗲 𝗮 𝗱𝗶𝗿𝗲𝗰𝘁𝗼𝗿: "slow this part down by half", "cut right here", "zoom in on the button".

The black hole lab quoted below was made by Opus 5.5 too: one prompt for the first version, then I went to sleep while a few agents took turns finding and fixing problems, and woke up to three rounds done. My short animation "small print" also came from Opus 5.5. My first prompt asked it to break down a reference video, and after rendering, it went back over the result second by second and polished it.

Send this prompt to Claude Code 👇

"Set up a motion video team in ~/.claude/agents that I can use for every product video. If HyperFrames isn't installed, install it first following https://t.co/FymJe8Xlht.

1. reference-breaker (effort: medium): use ffmpeg to pull one frame per second from the reference video I give you, tile them into one image to look at, and break down the pacing, type, colors and transitions into a shot list plus HyperFrames starter code. Break down only; don't build.
2. storyboarder (effort: medium): from the shot list, make 3 storyboard options with a few key frames each, tiled into one image for me to pick from. After I pick one, use npx hyperframes snapshot to make one still per scene.
3. The main session builds the video: write each scene in HyperFrames, and if there's no music, synthesize a track with Python. Render a draft with npx hyperframes render --quality draft first, and render the final only after I approve.
4. film-reviewer (effort: high): use ffmpeg to pull one frame per second and review it second by second. List problems as P0, P1 and P2 (P0 must be fixed). Review only; don't edit. The main session fixes every P0 before showing me.

Every subagent needs Bash and must be able to view images. Before starting, ask me for the reference video, logo, colors, fonts and product screenshots. Show me the contents of the files you'll create, then explain the flow in words a 10-year-old could follow. Don't write anything until I confirm."

[herbst @hybridherbst]

What if... rigging and skinning were effortless

Solving skin weights, IK chains, deformation, floor contacts, volumes, … on the GPU, with @threejs

[Dilum Sanjaya @DilumSanjaya]

Tried turning a Dune sandworm into a robot concept with Opus 5.5

[Pounds @pounddz]

Blender + Seedance 2.5 is CRACKED for AI animations

I made this in a couple hours (not including scripting) and blender made it so much easier to direct the shots in the way I wanted

+Made it 3x cheaper to make because I had to do less regens

I'm leaking my whole workflow to make these animation ads super high quality like this

Here's what you're getting:
>Character design workflow
>Story board workflow
>How I create my videos for as cheap as I can
>My blender workflow
>Prompt structure
>How I ideate and write my copy
And more

if you want it reply under this + RT this and I'll send it over (gotta be following so I can DM)

[Maivar @RadarDiplo]

@Simon11726406 @Zirconicav @EmmanuelRuimy 🫥

[Maivar @RadarDiplo]

@Simon11726406 @Zirconicav @EmmanuelRuimy 1) ils en ont pas perdu le controle ça c'est la belle historie que les medias mainstream te racontent
2) mais mec en gros tu penses que assad massacrait des civils par plaisir et que les USA sont venus gentiment après donner des armes a des gentils révolutionnaires c'est ça ?

[Fabien @Fabien_Mikol]

Un modèle de langage généraliste (en l'occurrence Astra), sans même d'entraînement spécifique antérieur, est capable de servir de "cerveau" à un robot humanoïde : explorer un environnement nouveau (une cuisine), s'en forger un modèle, et résoudre des tâches même longues ou vagues

[Maivar @RadarDiplo]

@Simon11726406 @Zirconicav @EmmanuelRuimy les attaques chimiques c'est encore plus délirant regarde l'artique de seymour hersch dessus

[Maivar @RadarDiplo]

@Simon11726406 @Zirconicav @EmmanuelRuimy bah oui évidemment ils vont crier sur tous les toits qu'ils financent le terrorisme. sur ce genre d'affaire sensible, tu imagines quoi ? qu'ils vont faire une conférence de presse en disant : Oui on finance des terroriste pour virer Assad qu'on aime pas ?

[INSIDER @x_insider4]

5 YEARS FOR NOTHING

I GAVE CLAUDE $100, A 410-MARKET WATCHLIST, AND NO INPUT FROM ME. IT OUTTRADED ME ON DAY ONE. $1,734 AT HOUR 24.

Claude Opus 5.5 started with $100. Not $1,000. Not $10,000. One hundred dollars.

It scanned 410 markets. Crypto, gold, silver. Then it traded only one at a time: whichever had the deepest order book right now.

Every pick got a score before entry. On screen they ranged from 0.87 to 0.96 out of 1. The other 402 markets stayed on the watchlist.

First pick: short Bitcoin at $84,538. Balance $103.28.

Then the machine kept rotating.

Solana. Ethereum. Gold. $128. $154. $179. $206.

XRP. Dogecoin. BNB. $305. $377. $515. $632. $931. $1,102. The curve went vertical.

Last pick of the day: silver. Hour 24: $1,734. From $100. 43 trades. 72.1% win rate.

It lost more than 1 trade in 4. It did not matter. Every market it picked had a 1 tick spread, so no slippage was eating the edge.

Most people pick a coin first and look for a reason second. Claude picked the market last, after reading 410 of them.

$1,634 profit in one day. A week like that is $11,438. From the price of a dinner for two.

Not once did a human choose a coin, set a stop or click buy. Claude ran all 24 hours alone.

The full setup is below and free. Save it before the next scroll buries it.

[Simplifying AI @simplifyinAI]

MICROSOFT JUST OPEN-SOURCED SELF-EVOLVING AGENT SKILLS.

it's called skillopt.

skills that improve themselves the same way you train an ai model.

no more guessing what changes will actually help your prompts.

the base model runs the task, then an optimizer evaluates the output and rewrites the instructions itself.

→ seperates successful paths from failures to find actual improvements
→ auto-rejects any edit that doesn't exceed your benchmark score
→ beats hand-crafted prompts and optimizers
→ zero model lock-in, the skill transfers to any model

100% free. open source.

[beamnxw ./ @beamnxw]

this is pure f*cking treasure

20 useful GitHub projects for a system that captures sources, connects ideas, remembers context, and helps agents act on it

your second brain should get sharper after every session

CAPTURE THE INPUT
01 Chubby Skills
▸ https://t.co/InxyRwMZtz
02 OpenWiki
▸ https://t.co/bLeUouBLuu
03 Agent Second Brain
▸ https://t.co/WANfN9tVP4
04 DocMason
▸ https://t.co/rQsMrIDGh0
05 DocsAgent
▸ https://t.co/R8DF9NI0tt

CONNECT THE IDEAS
06 claude-obsidian
▸ https://t.co/uhwe1f5n6X
07 SwarmVault
▸ https://t.co/QNxk0QvLYH
08 sage-wiki
▸ https://t.co/OQIkSucHAA
09 Vault Curate
▸ https://t.co/FeR1r3vLle
10 COG Second Brain
▸ https://t.co/1gtpmSyJuy

REMEMBER WHAT MATTERS
11 Hindsight
▸ https://t.co/3cXNIFAyDN
12 memU
▸ https://t.co/iCoOwoBIT3
13 TencentDB Agent Memory
▸ https://t.co/i2DaJnwGhu
14 agentmemory
▸ https://t.co/3OlbQk5CPq
15 OpenViking
▸ https://t.co/Ydr8K54XxW

PUT IT TO WORK
16 Open Second Brain
▸ https://t.co/DiOBxmK4Gh
17 Second Brain Cloudflare
▸ https://t.co/WlokLe12JV
18 makerskills
▸ https://t.co/Q1RQFhStsL
19 Row-Bot
▸ https://t.co/I1KBQwA7hk
20 MateClaw
▸ https://t.co/N9vGfry4no

the loop:

capture the source → turn it into linked, checkable knowledge → recall the right context → act with it → write the correction back

3 builds I'd explore:

creator: Chubby Skills → SwarmVault → Hindsight → makerskills
researcher: DocsAgent → sage-wiki → memU → Row-Bot
team: DocMason → COG Second Brain → TencentDB Agent Memory → MateClaw

save this, then build a second brain for your business ⭣

[Money Mike @zac_clarke63]

Over the next five years, biotech stocks will generate immense, generational wealth for investors.

Here are five stocks with the potential to grow fivefold:

1) $HIMS