Views
No views yet
Our Loopy generates high-quality looping videos with seamless transitions at loop boundaries and diverse motion. It also supports RGBA with semi-transparent effects. In the application block, all elements—including game assets and Loopy character stickers—are generated by our Loopy.
| Preview Video | Preview Video |
|---|---|
1# Clone the project repository
2git clone https://github.com/WeChatCV/Loopy.git
3cd Loopy
4
5# Create and activate Conda environment
6conda create -n Loopy python=3.11 -y
7conda activate Loopy
8
9# Install dependencies
10pip install -r requirements.txt1output_path="./checkpoints"
2mkdir -p $output_path
3
4torchrun --nproc_per_node=8 generate_2.2_new.py --task t2v-A14B --size 832*480 \
5 --ckpt_dir Wan-AI/Wan2.2-T2V-A14B \
6 --dit_fsdp --t5_fsdp --ulysses_size 8 \
7 --frame_num 53 \
8 --sample_steps 4 \
9 --high_lightx2v_path "wan2.2_lora/wan2.2_t2v_A14b_high_noise_lora_rank64_lightx2v_4step_1217.safetensors" \
10 --low_lightx2v_path "wan2.2_lora/wan2.2_t2v_A14b_low_noise_lora_rank64_lightx2v_4step_1217.safetensors" \
11 --high_lora_path high_noise.safetensors \
12 --low_lora_path low_noise.safetensors \
13 --prompt_file prompt.txt \
14 --output_dir $output_path/results 2>&1 | tee -a $output_path/results.log
15bash test.shWan2.2-T2V-A14B with --ckpt_dir, the LightX2V distillation
LoRAs with --high_lightx2v_path / --low_lightx2v_path, and the Loopy LoRAs with
--high_lora_path / --low_lora_path. Wan2.2-T2V-A14B is a two-expert MoE model, so the
high-noise and low-noise branches each need their own pair of LoRAs.--output_dir, one subdirectory per prompt, each containing
fgr.mp4.1# An example of prompt.
2An Arctic fox leaps nimbly through the snow while hunting, its white fur blending seamlessly with the snowflakes. Realistic style; a scene of winter wildlife.1@article{haotiandong2026loopy,
2 title = {Loopy: Seamless Video Loop Generation via Anchored Looping Shift of Positional Embedding},
3 author = {Haotian Dong, Wenjing Wang, Chen Li, Jing Lyu, Xin Wang, Di Lin},
4 journal = {arXiv preprint},
5 year = {2026}
6}