Open-Source Projects

LiveAvatar

Real-time, streaming, infinite-length audio-driven avatar generation powered by a 14B diffusion model — 45 FPS on 5×H800, FP8 release runs on 48GB GPUs. ECCV 2026. HF #1 Paper of the Day, 2k+ stars.

Code Project Paper

EasyControl

Efficient and flexible condition-guided control for Diffusion Transformers (DiT). Plug-and-play Condition Injection LoRA, position-aware training, and KV-Cache-accelerated inference. ICCV 2025.

Code Project Paper Demo

EasyControl Ghibli

One-click Ghibli-style portrait generation built on EasyControl. Ranked the #1 trending Space on Hugging Face.

Demo

PhotoDoodle

The first in-context architecture for image editing & customizable image editing.

StableMakeup

Robust real-world makeup transfer with diffusion models — from light to extremely heavy styles. SIGGRAPH 2025.

Code Paper

Stable-Hair

Real-world hair transfer via diffusion models, robustly handling diverse and complex hairstyles. AAAI 2025; V2 in IEEE TVCG 2026.

Paper

SSR-Encoder

Encoding selective subject representation for subject-driven generation — high-fidelity, query-guided personalization without test-time fine-tuning. CVPR 2024.

Project Paper