KoLτ-Real 🛡️

Another multi-image to 3D mesh model, but this one can also gen PBR materials. Can't really use it for generated images, since genned images don't have great consistency, but it could be neat for real-life objects.

🚨 BREAKING: NVIDIA just removed the biggest friction point in Voice AI They open-sourced PersonaPlex 7B, a real-time conversational model. It listens and speaks simultaneously to handle natural interruptions and overlaps. 100% Open Source.

GaussianGPT. Autoregressive 3D Gaussian scene generation. - XYZ voxel serialization - Lookup-free quantization (LFQ) - Infinite chunk-based outpainting ideal for procedural gaming/VR level design. nicolasvonluetzow.github.io/GaussianGPT/

Netflix dropped some useful stuff. VOID -video object and interaction deletion. - removes objects while realistically simulating physical consequences; - beats Runway/ProPainter; - CogVideoX-5B + SAM 2; looks good, no smudges/artifacts void-model.github.io

🚨 Investir dans Mistral AI avant son introduction en bourse est maintenant possible. Mais voilà ce qu'il faut vraiment comprendre avant de se lancer. Depuis le 1er avril 2026, une offre secondaire sur Crowdcube permet aux particuliers d'accéder indirectement au capital de Mistral AI à une valorisation de 11,7 milliards d'euros (14 milliards de $). Mistral, c'est quoi concrètement ? 🔸 La startup IA la plus valorisée d'Europe 🔸 Fondée en 2023 et déjà 4e licorne IA mondiale 🔸 Objectif d'1 milliard d'€ de revenus en 2026 Pour contextualiser OpenAI vient de lever 122 milliards de dollars à une valorisation de 852 milliards. Mistral est valorisé 60 fois moins pour un positionnement radicalement différent : la souveraineté numérique européenne. Ses modèles open-weight se déploient directement sur les serveurs de l'entreprise cliente. 0 donnée qui transite vers des infrastructures étrangères. C'est son argument clé auprès des banques, hôpitaux, et institutions publiques. Mais avant d'investir, il faut comprendre ce que vous achetez concrètement. Vous achetez des parts d'un SPV britannique, qui détient des parts dans une entité luxembourgeoise, qui elle-même détient des actions Mistral AI. Des actions Seed et Série A, les premières émises qui sont moins bien protégées. En cas de liquidation, les détenteurs de Série C (ASML, Nvidia…) sont remboursés en priorité. Quelques informations complémentaires : - 5 % de frais d'entrée - 5 % de carried interest sur les plus-values - Pas de droit de vote - Ticket d'entrée : 107,80 € minimum (26,95 € / action) Est-ce que la thèse "souveraineté numérique européenne" justifie une valorisation de 14 milliards de dollars, sachant qu'OpenAI, qui n'est toujours pas profitable, en vaut 852 ?

Wait… what!?

ComfyUI Kimodo, build new custom node for Kimodo inside ComfyUI

Just open sourced this ComfyUI 3D Architecture Pascal Editor

PackForcing. Wan2.1 high-quality long video gen up to 120s. - train on 5s clips for 24x temporal extrapolation; - reduces KV-cache from 138GB to 4GB; - tops VBench long-video benchmarks github.com/ShandaAI/PackF…

Cool. A local, free Topaz+NanoBanana alternative, almost StepFun just dropped a real gift: RealRestorer, a fully open-source image restoration model. - removes rain, fog, glare - no more moiré, JPEG artifacts - restore low-light - 1024×1024 based on Step1X-Edit DiT + Flux-VAE yfyang007.github.io/RealRestorer/

Testing the new LTX 2.3 IC LORA for head swaps. Pretty good straight out of the box! Will have to play more with this haha Link below

ComfyUI-SegviGen for precise 3D part segmentation. Auto model downloading (Trellis, SegviGen, BiRefNet, DinoV3). Post-processing (VXZ, Latent Slats, Voxel, GLB). github.com/Aero-Ex/ComfyU…

🚨BREAKING: Someone just open-sourced a headless browser that runs 11x faster than Chrome and uses 9x less memory. It's called Lightpanda and it's built from scratch specifically for AI agents, scraping, and automation. Not a Chromium fork. Not a hack. A completely new browser written in Zig. Here's why this changes everything for AI builders: ↓

FLUX.2 [klein] 9B just got 2x faster at image editing, especially when you use multiple reference images. Same quality, no price increase.

Another neato node for LTX2 2.3. I'm honestly on the fence to say that it's clearly better or worse though. Saturation did indeed get bumped up though.

These are the MOST SOLID subnets in the entire ecosystem right now 🔥 Strongest ideas + battle-tested teams = real alpha. No fluff, just builders. templar #SN3 > affine #SN120… every single one on this list is built different. Newbies, this is your easiest way to start strong 😎 #Bittensor #TAO $TAO

Anthropic released a complete playbook on "How to build a Polymarket arbitrage bot" They literally explained how to create your own SKILL, which can arbitrage, trade or do anything else for you. Those who read this file, will win those who didn't. Link below