🧵 PractiLight: Practical Light Control Using Foundational Diffusion Models 🔗 https://t.co/YweM7ir948 Diffusion models already “know” a lot about light transport. No need for massive finetuning that hurts generalization. {1/6} 👇

if you are looking to understand the intuition behind diffusion models check out @juliarturc most recent video amazing work and spendid visual highly recommend

.@yupp_ai compared 700+ image models, and the results are clear. On the preference side, @GoogleDeepMind’s Gemini 2.5 Flash Image (aka “Nano Banana”) takes the crown, with @OpenAI’s GPT Image 1 close behind. On the speed front, @StabilityAI’s Stable Diffusion 3.5 Large Turbo

In deep learning, diffusion models are a type of generative model that creates new data (like images) by reversing a process of adding noise. They work in two main stages: a forward diffusion process that progressively adds Gaussian noise to an image until it becomes pure noise,

New survey on diffusion language models: https://t.co/SHicf69gxV (via @NicolasPerezNi1). Covers pre/post-training, inference and multimodality, with very nice illustrations. I can't help but feel a bit wistful about the apparent extinction of the continuous approach after 2023🥲

Does a smaller latent space lead to worse generation in latent diffusion models? Not necessarily! We show that LDMs are extremely robust to a wide range of compression rates (10-1000x) in the context of physics emulation. We got lost in latent space. Join us 👇

Dayani et al., "MV-RAG: Retrieval Augmented Multiview Diffusion" Retrieval augmentation for 3D recon with multi-view diffusion models. Trains both with 3D and 2D assets during training, and uses 2D natural images for inference. Makes sense, I guess? Many things look alike!

Intelligence isn’t about bigger models, it's about smarter modules. OpenVision introduces Adapter Diffusion, a system where custom-trained adapters spread across the network like open-source plug-ins. Instead of retraining or hardcoding, adapters: Integrate seamlessly into the

Captivated by the intricate beauty of this AI-generated mosaic! ✨ This piece, inspired by African artistry, shows the incredible power of generative AI and diffusion models to create stunning, culturally rich visuals. #GenerativeAI #DiffusionModels #AfricanArt #GlobalCulture

Neat paper on linear classifier-free guidance: https://t.co/0sTuJgO330 I think that the theoretical tractability of diffusion models, and the unique forms of interpretability that that enables, continue to be quite under appreciated.

Excited to share our work Set Block Decoding! A new paradigm combining next-token-prediction and masked (or discrete diffusion) models, allowing parallel decoding without any architectural changes and with exact KV cache. Arguably one of the simplest ways to accelerate LLMs!

Happy to share a side project: Diffusion Language Models Know the Answer Before Decoding. Diffusion LMs are often dismissed as slow. But what if they already *know* the answer halfway through? 1. Early Answer Convergence: Our new paper shows that in many cases, they do,

Last month I presented our work on “Inference-Time Guided Generation w/ Diffusion & Flow Models” at NVIDIA, Google, Stanford, and SFU. I showed how recent flow matching models can be especially powerful for inference-time guidance. Check out the slides: https://t.co/1RaszIV2Ar

🚀 Looking to connect with people interested in: • Web Development • JavaScript • Automation • AI • ML • Web3 • Generative AI (LLMs, Diffusion Models, AI Art) • MERN • FastAPI • Next.js / Svelte / SolidJS • DBMS (SQL / NoSQL / NewSQL)

🚀 Looking to connect with people interested in: • Web Dev • JavaScript • Automation • AI • ML • Web3 • Generative AI (LLMs, Diffusion Models, AI Art) • MERN • FastAPI • Next.js / Svelte / SolidJS • DBMS (SQL / NoSQL / NewSQL)

Revolutionizing Reinforcement Learning Framework for Diffusion Large Language Models

Congratulations and farewell to my Masters student Jimmy Z. Di (@jimmy_di98). He presented his thesis on memorization in diffusion models, in addition to his other great unlearning work. He goes on to a PhD at @UWMadison. We will miss you!

Robots don’t just need to move ,they need to understand. That’s why the robotics world is converging on VLA models: Vision–Language–Action. A VLA combines a large language model (to interpret instructions and images) with a diffusion model (to generate the next sequence of

🌟 Excited to share our #ICCV2025 paper: Make Me Happier: Evoking Emotions Through Image Diffusion Models 🎨😊We show how diffusion models can be guided to generate images that evoke targeted emotions --- emotional AI. 📄 Paper & 💻 Code: https://t.co/z9E4SJCuIp