Train a real-time, playable Video World Models on 8 GPUs — keyboard & mouse control & gamepad, fully open and reproducible.
Official code, locked protocols, and audit artifacts for "Jev-Routed Temporal Granularity with Recovery and Hysteresis for Causal Video Generation": per-segment fine/coarse/hold routing for Causal-rCM (Wan2.1-1.3B) via the Jev oracle, with two-GPU (T4/A100) feasibility audits.
Quartz — custom CUDA/Vulkan GGUF inference engine + SD1.5/SDXL image and video generation (Rust, no llama.cpp)
a careful thought out approach to installing ComfyUI with Wan2.1, manager and some other specialty nodes that are a must on day one for me.
Unofficial ComfyUI integration of Causal Forcing / Causal Forcing++ (official frame-wise 2-step and 4-step checkpoints, official inference path, separate WSL2/Linux runtime)
Make AI trailers 100% locally on a 6 GB GPU: video (Wan 2.1), music (ACE-Step), narration (OmniVoice) and ffmpeg mix, driven by a storyboard.json, by hand or by an LLM agent.
Unofficial ComfyUI integration of MonarchRT (training-free, on the public Self-Forcing / Wan2.1-T2V-1.3B weights) with a dense A/B baseline, run in a separate WSL2/Linux runtime.
Unofficial ComfyUI nodes for DuoMatching streaming text-to-video (Wan2.1-T2V-1.3B, 2 steps per frame); in-process, no flash-attn; bit-identical to the official script
InfiniteTalk (Wan2.1) talking-avatar pipeline: ComfyUI workflows, RunPod pod/serverless setup, A/B matrix and forensic evaluation vs HeyGen
InfernoBench: open-source AI model installers, local inference benchmarks, and reproducible browser labs. Wan2.1 and Sulphur-2 included.
Self Gradient Forcing+ for ComfyUI: autoregressive text-to-video with the official SGF+ models (Wan2.1-T2V-1.3B), in-process, bit-identical to the official script
Local text to video pipeline that turns a text prompt into a lip synced talking video. Wan2.1 handles video generation, Qwen3 TTS creates the voice, and MuseTalk 1.5 syncs the lips. Everything runs offline after the models are downloaded.
Streaming video diffusion on one AWS Trainium2 chip. StreamDiffusionV2’s causal Wan2.1 1.3B model ported to Neuron with a resident KV cache and a 4-NeuronCore layer pipeline: 5.6 → 38.4 FPS end to end, matching the CPU reference (cosine 0.99996). Plus a live SD-Turbo webcam restyle at 43.8 FPS.
ICDepth: Taming Video Diffusion Models for Video Depth Estimation via In-Context Conditioning (ECCV 2026)
🎥 AI Video Generation Studio — Text-to-Video & Image-to-Video engine with Gradio & Free Cloud GPU Acceleration
To associate your repository with the wan2-1 topic, visit your repo's landing page and select "manage topics."