import torch
from diffusers import DiffusionPipeline
# switch to "mps" for apple devices
pipe = DiffusionPipeline.from_pretrained("sayakpaul/q8-ltx-video", dtype=torch.bfloat16, device_map="cuda")
prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k"
image = pipe(prompt).images[0]This is the Q8 version of the LTX-Video transformer (revision: 0c43a96).
This work largely builds on top of KONAKONA666/LTX-Video, utilitizing the Q8 kernels from KONAKONA666/q8_kernels. These kernels are optimized to run on the NVIDIA GPUs with ADA architecture (example RTX 4090). Huge shoutout to KONAKONA666 for working on these.
How does it differ from konakona/ltxvideo_q8?
The purpose of this checkpoint is to be more closely compatible with the official LTX-Video transformer shipped in diffusers.
Refer to this repository for more details on the checkpoint was obtained and how to use it. The repository also provides some benchmarks and here's a summary:
- Downloads last month
- 26
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support