Instructions to use madebyollin/taesd with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Diffusers
How to use madebyollin/taesd with Diffusers:
pip install -U diffusers transformers accelerate
import torch from diffusers import DiffusionPipeline # switch to "mps" for apple devices pipe = DiffusionPipeline.from_pretrained("madebyollin/taesd", dtype=torch.bfloat16, device_map="cuda") prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k" image = pipe(prompt).images[0] - Notebooks
- Google Colab
- Kaggle
Eyes problem
#2
by recoilme - opened
Thanks for the report, will take a look. Version 1.2 should improve texture detail but there might be other regressions to address in follow-ups. If it helps, you can switch back to 1.1 by downloading weights from https://huggingface.co/madebyollin/taesd/tree/73b7d1c8836d16997316fb94c4b36a9095442c1d (or setting revision=73b7d1c8836d16997316fb94c4b36a9095442c1d in from_pretrained).
Update: the new version (1.3) should have slightly better quality than 1.2 or 1.1. I also saw a retrained checkpoint from ByteDance here https://huggingface.co/cqyan/hybrid-sd-tinyvae which you could try using. sd-vae-ft-ema is still the gold standard OFC :)

