Wan
Also known as: Wan 2, Wan2
13stories this week
30last 30 days
31all time
Timeline
- Oct 19, 2026 · Opinion / analysis · 1 sourceLTX 2.5 running locally on a 48GB M5 Pro MacBook Pro in TensorSharp’s Video StudioI maintain TensorSharp, an open-source local inference engine and agent runtime.
- Oct 19, 2026 · Open-source release · 1 sourcesgl-project/sglang v0.5.22| Model | Type | Cookbook |
- Oct 18, 2026 · Opinion / analysis · 1 sourceKroma vs Moss, is it just me?I've seen so many high-quality Moss images posted here but when I try Moss on my PC, the img quality is absolutely horrible + render time.
- Oct 17, 2026 · Opinion / analysis · 1 sourceRecap on Best Open Source AI Models so farSo after so many models have been released the past year, what are everyone's top models so far?
- Oct 16, 2026 · Research paper · 1 sourceSink-Sparse Hybrid Attention for Efficient Inference in T2I/T2V Diffusion TransformersDiffusion Transformers (DiTs) have become a common backbone for high-quality text-to-image and text-to-video generation, but repeated dense self-attention remains a major inference bottleneck as token sequences grow and the same operator is invoked across many denoising steps.
- Oct 15, 2026 · Research paper · 1 sourceSoftCache: Adaptive Step Size Caching for Diffusion TransformersDiffusion and flow models generate a sample over iterative steps, each a full pass through a large network, and that pass dominates generation cost.
- Oct 15, 2026 · Research paper · 1 sourceWhat Should Robot Policies Learn to Predict? Rethinking Supervision Targets in Vision-Language-Action LearningVision-language-action (VLA) policies are typically trained with action prediction as the primary supervision signal, since robots ultimately execute actions.
- Oct 15, 2026 · Research paper · 1 sourceSparse Attention for Video Generation Acceleration via Growing Sparsity and Reduced SearchBased on these findings, we propose Growing Sparsity and Reduced Search (GSRS), a training-free framework that retains more computation early and progressively reduces it later, while restricting subsequent searches to an early candidate set to reduce search overhead.
- Oct 14, 2026 · Opinion / analysis · 1 source20 steps, 720p, RTX 3090: A narrative video experiment with Wan2GPI've been experimenting with Wan2GP to see if I could make something closer to an actual TV episode instead of the usual collection of unrelated AI clips.
- Oct 13, 2026 · Research paper · 1 sourceDuplexBridge: Coordinating Understanding and Generation for Unified Image EditingWe present DuplexBridge, which uses a unified model's own understanding capability to enhance prompts for its own generation capability, without an external understanding module.
- Oct 13, 2026 · Research paper · 1 sourceGAD: Geometry Attention Distillation for Consistent Video GenerationThus, we introduce Geometry Attention Distillation (GAD), which transfers correspondences from a frozen geometry model's attention to video attention.
- Oct 13, 2026 · Research paper · 1 sourceRSIgen: Reciprocal Self-Improvement for Agentic Visual GenerationAgentic visual generation produces experience that can improve the native understanding (agent) and generation roles of a unified multimodal backbone, but task success alone does not determine which role should learn.
- Oct 13, 2026 · Open-source release · 1 sourceunslothai/unsloth v0.1.907-beta: Large Performance Gains + More Browser UseUnsloth has gotten major performance upgrades!
- Oct 10, 2026 · Research paper · 2 sourcesLightVAE: Towards Compact and Efficient Video AutoencodersTo address these limitations, we present LightVAE, a compression framework that exploits redundancy in temporal modeling and network depth.
- Oct 9, 2026 · Research paper · 1 sourceGraftOOD: Grafting Out-of-Distribution Exemplars into Advantage-Weighted Rectification for Video Reward OptimizationVideo reward optimization can increase proxy rewards while reducing video dynamics, a mismatch we identify as motion collapse, including under a video-level motion-quality reward.
- Oct 9, 2026 · Research paper · 1 sourceVideoAWR: Reinforcing Video Flow Models with Decoupled-Timestep Advantage-Weighted RectificationRegression-based RL for flow models mandates sampling without classifier-free guidance (CFG) to align the rectification target with the reference marginal velocity.
- Oct 8, 2026 · Research paper · 1 sourceWorldAlign: Decoupled 4D Reward for World-Consistent Video GenerationFaithful visual world simulation requires generated videos to maintain 4D world consistency, encompassing both static and dynamic consistency.
- Oct 8, 2026 · Open-source release · 1 sourceunslothai/unsloth v0.1.905-beta: Sandboxing is here!We're introducing Windows, Mac and Linux sandboxing in Unsloth!
- Oct 8, 2026 · Research paper · 1 sourceMemory Forcing: Attendable Mid-Horizon History for Streaming Video GenerationAutoregressive video diffusion enables causal video streaming without a bidirectional pass over the full clip, but existing few-step systems usually retain only the opening and most recent frames in a fixed-size KV cache.
- Oct 8, 2026 · Research paper · 1 sourceTowards Unified Evaluation of Prompt Enhancers for Video GenerationTo address this gap, we introduce PEBench, the first unified benchmark for direct PE evaluation across text-to-video, image-to-video, and reference-to-video prompt enhancement.
- Oct 8, 2026 · Research paper · 1 sourceParametric Trajectory Distillation for Few-Step Video GenerationWe introduce Parametric Trajectory Distillation (PTD), which lets the student parameterize teacher trajectory segments as polynomials and learn from teacher guidance along its own predicted path.
- Oct 8, 2026 · Research paper · 1 sourceiCATS: Fast Video Generation via Interaction-Aware Sparse Attention and Timestep-Adaptive SparsityTraining-free sparse attention offers a practical acceleration solution to Diffusion Transformers (DiTs) via reducing computations without fine-tuning.
- Oct 7, 2026 · Research paper · 1 sourceFluid-Gen-Zero: Grounding Pretrained Video Generators in Physics without TrainingWe present Fluid-Gen-Zero, a training-free framework for physics-aware fluid-object interaction video generation that decouples physical reasoning from appearance synthesis.
- Oct 7, 2026 · Research paper · 1 sourceGRACE: Generation-aware latent compression for efficient video generationTo address this, we propose Generation-Aware Latent Compression for Efficient Video Generation (GRACE), a two-stage framework that compresses a pretrained video autoencoder while keeping it compatible with the pretrained DiT.
- Oct 7, 2026 · Open-source release · 1 sourceunslothai/unsloth v0.1.904-beta: Train your own Decision modelTurn any text or vision LLM into a Jev-style decision model in Unsloth, with decision accuracy going from 30% to 80%.
- Oct 6, 2026 · Research paper · 1 sourceBackend-Agnostic Sparse Attention for Fast High-Resolution Visual GenerationTo tackle these challenges, we propose BASA, a backend-agnostic sparse attention, which brings the best of both worlds: visual quality and practical acceleration.
- Oct 6, 2026 · Research paper · 1 sourceOpenWAM: An Open Framework for Composable World-Action ModelsWe introduce OPENWAM, an open world-action modeling framework built around a common causal robot-video foundation and configurable video-action interaction.
- Oct 6, 2026 · Open-source release · 1 sourcespicyapi-ai/comfyui-spicyapi: ComfyUI custom nodes for 180+ hosted AI models: Seedance, Kling, Wan, MiniMax H3, Vidu, LTX, Suno, GPT Image, Nano Banana, Seedream, FLUX LoComfyUI custom nodes for 180+ hosted AI models: Seedance, Kling, Wan, MiniMax H3, Vidu, LTX, Suno, GPT Image, Nano Banana, Seedream, FLUX LoRA, text to speech and more.
- Oct 6, 2026 · Open-source release · 1 sourcehuggingface/diffusers v0.41.0: Diffusers 0.41.0: QwenImage 2.1 pipeline and more> This release brings Qwen-Image 2.1 to Diffusers, with text-to-image generation, image editing, native transparency, and LoRA training.
- Sep 29, 2026 · Open-source release · 1 sourceComfy-Org/ComfyUI v0.38.0