Image-to-Video
Diffusers
Safetensors
world-model
video-generation
text-to-video
action-conditioned
causal
Instructions to use seedleap/zing-0.5 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Diffusers
How to use seedleap/zing-0.5 with Diffusers:
pip install -U diffusers transformers accelerate
import torch from diffusers import DiffusionPipeline from diffusers.utils import load_image, export_to_video # switch to "mps" for apple devices pipe = DiffusionPipeline.from_pretrained("seedleap/zing-0.5", dtype=torch.bfloat16, device_map="cuda") pipe.to("cuda") prompt = "A man with short gray hair plays a red electric guitar." image = load_image( "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/guitar-man.png" ) output = pipe(image=image, prompt=prompt).frames[0] export_to_video(output, "output.mp4") - Notebooks
- Google Colab
- Kaggle
Questions on model details.
#2
by yunming181920 - opened
I noticed you mentioned the model is 5B. Is it trained based on Wan2.2 5B? Additionally, on a single RTX 5090, what FPS could be achieved with Sage3+SLA enabled? Compared to other mature video-based world models, what novel innovations does this model offer?
I noticed you mentioned the model is 5B. Is it trained based on Wan2.2 5B? Additionally, on a single RTX 5090, what FPS could be achieved with Sage3+SLA enabled? Compared to other mature video-based world models, what novel innovations does this model offer?
Thank you for your interest!
- Yes, our model is trained based on Wan2.2-5B.
- On a single RTX 5090, it can generate 480p streaming video at over 24 FPS. We also plan to open-source our SGLang-based inference code.
- Compared with existing video world models, our model supports joint control through both text prompts and WASD actions, and achieves strong results in prompt-guided, action-controllable video generation.