Ervilha

[AI Search @aisearchio]

SolarWM turns Minimax H3 into realtime interactive worlds. It can keep generating for over an hour.

https://t.co/uBEipIiL2h

[Xingyi Yang @yxy2168]

We share H3-World 🌍
The first to turn MiniMax-H3 itself into world model.

No new action module. We directly convert H3’s pretrained language understanding into world control.

Only 8K samples + 0.199% Trainable Params.

📄 https://t.co/Bn4M4JwEJs
💻 https://t.co/sT44pUHAl8

[AI Search @aisearchio]

MiniMax Design is here! It's a multimodal agent + infinite canvas app, which allows you to quickly create and iterate content.

It can also work with your local files, read web pages, and connect with editing tools.

One of my favorite strengths of Minimax is that it's able to preserve references well. Notice it's able to get the tiny text of the product consistent across all videos.

https://t.co/qCpfJgnrBE

[AI Search @aisearchio]

Controlnet for Minimax H3 is here!

https://t.co/ffRQguhmGx

[Perturabo, Lord of Iron @RedeemerOfNod]

This might be the funniest ‘minor spelling mistake’ gif I’ve ever seen

[いっぬしか勝たん🐕_PR @bolootaa]

@actionecomics Tem q ser meme… Tolkien deve estar energizando o bairro td de tanto girar dentro do túmulo

[toyxyz @toyxyz3]

MiniMax-H3 Latent Upscaler T2V test #comfyui

[AI Search @aisearchio]

An experimental node pack that brings SPEED (Spectral Progressive Diffusion) to MiniMax-H3.

It starts denoising at low resolution (cheap and fast), then gradually ups the resolution. You get similar quality with less VRAM and faster results. Audio stays at full resolution the whole time.

https://t.co/ffuYT4RDnD

[AI Search @aisearchio]

Here's a way to make long continous video with Minimax H3.

This workflow stitches clips into seamless long videos with consistent motion, visuals, and audio. Much friendlier than the other workflow I shared last week.

https://t.co/VkKMvS94uW

[AI Search @aisearchio]

This is a neat way to add video effects in MiniMax H3 without loras.

They mix together and can be initialized or trained from prompts, images, or video. An alternative to lora training.

https://t.co/NDs770vlYH

[AI Search @aisearchio]

This Minimax H3 workflow is WAY easier. Handles t2v, i2v, r2v all in one simple workflow.

ComfyUI-MiniMaxH3-Easy
https://t.co/IKPW6hIIqK

[AI Search @aisearchio]

FlaRe is a floating radiance network that combines editable geometry with continuous neural radiance for ray-traceable scenes.

It supports rendering, recursive ray tracing, deformation, stylization, and mesh extraction in one compact representation.

https://t.co/4GskXayWWo

[AI Search @aisearchio]

Qwen Video Edit adapts Qwen Image Edit to do video editing.

It edits video-VAE latents directly, avoiding a video pretrained transformer while supporting long and portrait clips.

https://t.co/JcnNSVrTAJ

[AI Search @aisearchio]

Very efficient Minimax H3 upscaler.

Generate in low resolution > upscale in latent space > vae decode

https://t.co/VlAXjDLQkr

[Dan Greenheck @dangreenheck]

Cooking up the Three.js Water Pro v3.4 update! 👨🏻‍🍳

You can now choose between physically-based Jerlov water types which model absorption based on algae, silt and strain, or define your own custom colors if you require more stylized rendering!

Releasing today or tomorrow.

[Insact @Insact9704]

red🐰

[𝗕𝗮𝗻𝗱𝗶𝗻𝗼𝗽𝗹𝗮 @bandinopla]

Introducing TSL @threejs Procedural Noise Functions Collection, port from GLSL by user "lumiey" on @XorDev's https://t.co/NBwgCh5d5S did (the ai did 😋) some modifications to make them seamless. Useful for anything related to procedural generation, random scatter, etc. link below

[Fabio Ottaviani @supahfunk]

Smoke wants out. Glass won’t let it.

#threejs #glsl #shaders

https://t.co/0JAUHbVzQj

[Patrick Jaillet @JailletPatrick]

https://t.co/hVZ7ED9pKt
vec2 d=(FC.xy-.5*r.xy)/r.y,L=d*1.5-vec2(.5,0.),a=vec2(0.);float t_0=t,s=1.+sin(t_0*.5)*.5,I=sin(t_0*.5),J=cos(t_0*.3),K=length(d),M=0.;vec3 e=vec3(0.);for(int u=0;u<16;u++){if(dot(a,a)>4.)break;a=vec2(a.x*a.x-a.y*a.y,2.*a.x*a.y)+L;++M;}vec2 n=a;float h=abs(n.y+I*.7)*abs(n.x+J*.7);h=pow(h,.45);vec2 v=n*3.5+t_0*.2,i=floor(v),c=fract(v);c=c*c*(3.-2.*c);float N=fract(sin(dot(i,vec2(127.1,311.7)))*4.3758547e4),O=fract(sin(dot(i+vec2(1.,0.),vec2(127.1,311.7)))*4.3758547e4),P=fract(sin(dot(i+vec2(0.,1.),vec2(127.1,311.7)))*4.3758547e4),Q=fract(sin(dot(i+vec2(1.),vec2(127.1,311.7)))*4.3758547e4),R=mix(mix(N,O,c.x),mix(P,Q,c.x),c.y),w=.7+.3*sin(t_0*.8+K*1.5+R*2.),z=abs(s-2.)*.008,A=(.05/(h+.004+z))*w,B=(.09/(h+.025+z))*w,j=clamp(A+B,0.,3.);vec3 g=mix(vec3(.15,0.,0.),vec3(1.,.2,.02),smoothstep(0.,.5,j));g=mix(g,vec3(1.,.7,.1),smoothstep(.5,1.2,j)),g=mix(g,vec3(1.,.95,.6),smoothstep(1.2,2.2,j));vec2 C=d*1.5-vec2(0.,t_0*.4),k=floor(C),smh1=fract(C);smh1=smh1*smh1*(3.-2.*smh1);float D=mix(mix(fract(sin(dot(k,vec2(127.1,311.7)))*43758.5453),fract(sin(dot(k+vec2(1.,0.),vec2(127.1,311.7)))*43758.5453),smh1.x),mix(fract(sin(dot(k+vec2(0.,1.),vec2(127.1,311.7)))*43758.5453),fract(sin(dot(k+vec2(1.),vec2(127.1,311.7)))*43758.5453),smh1.x),smh1.y);vec2 E=d*3.-vec2(t_0*.1,t_0*.8),l=floor(E),smh2=fract(E);smh2=smh2*smh2*(3.-2.*smh2);float S=mix(mix(fract(sin(dot(l,vec2(127.1,311.7)))*43758.5453),fract(sin(dot(l+vec2(1.,0.),vec2(127.1,311.7)))*43758.5453),smh2.x),mix(fract(sin(dot(l+vec2(0.,1.),vec2(127.1,311.7)))*43758.5453),fract(sin(dot(l+vec2(1.),vec2(127.1,311.7)))*43758.5453),smh2.x),smh2.y),p=D*.6+S*.4,U=exp(-s*.12);vec3 T=mix(vec3(.05,.04,.04),vec3(.25,.18,.12),p);e+=g*(A+B)*U,e=mix(e,T*(1.2+j*.5),p*.65);float V=d.x*.8+d.y*.5-sin(t_0*.6)*1.2,F=exp(-V*V*12.);F*=(.4+.6*D);vec3 W=vec3(1.,.65,.3)*F*.8;e+=W*(1.-p*.5);vec3 q=e;q/=(1.+q*.2),o=vec4(q,1.);
#つぶやきGLSL, #GLSL, #shadertoy, #shaders, #creativecoding, #codegolf, #sizecoding, #twigl