AI agent skill that researches any topic across Reddit, X, YouTube, HN, Polymarket, and the web - then synthesizes a grounded summary

Domain-specific language designed to streamline the development of high-performance GPU/CPU/Accelerators kernels

AI agents running research on single-GPU nanochat training automatically

verl/HybridFlow: A Flexible and Efficient RL Post-Training Framework

A Lightweight LLM Post-Training Library

ComfyUI custom nodes for Diffusion Attentive Attribution Maps (DAAM)

Lets make video diffusion practical!

Fully automatic censorship removal for language models

①[ICLR2024 Spotlight] (GPT-4V/Gemini-Pro/Qwen-VL-Plus+16 OS MLLMs) A benchmark for multi-modality LLMs (MLLMs) on low-level vision and visual quality assessment.

SCAIL: Towards Studio-Grade Character Animation via In-Context Learning of 3D-Consistent Pose Representations (CVPR 2026 Findings)

for tile the image for advanced control or modification

Practice Code for text to image trainer

Qwen-Image text to image lora trainer

From baby GPT to diffusion GPT: An annotated implementation of a character-level discrete diffusion model (adapted from Karpathy’s baby GPT).

Visual Novel Character Creation Suite is a comprehensive tool for creating character sprites for visual novels. It allows you to create unique characters with a consistent appearance across all ima…

Tokamax: A GPU and TPU kernel library.

Official implementation of "Normalized Attention Guidance"

a Jax quantization library

Unofficial extension implementation of CausVid

[NeurIPS 2025] Sekai: A Video Dataset towards World Exploration

Official codebase for "Self Forcing: Bridging Training and Inference in Autoregressive Video Diffusion" (NeurIPS 2025 Spotlight)

A fast AI Video Generator for the GPU Poor. Supports Wan 2.1/2.2, LTX-2, Qwen Image, Hunyuan Video, LTX Video and Flux.

Official PyTorch implementation for "Large Language Diffusion Models"

Phantom: Subject-Consistent Video Generation via Cross-Modal Alignment

A TTS model capable of generating ultra-realistic dialogue in one pass.

SkyReels-V2: Infinite-length Film Generative model

A Distributed Attention Towards Linear Scalability for Ultra-Long Context, Heterogeneous Data Training