Instructions to use darkstorm2150/Protogen_Infinity_Official_Release with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Diffusers
How to use darkstorm2150/Protogen_Infinity_Official_Release with Diffusers:
pip install -U diffusers transformers accelerate
import torch from diffusers import DiffusionPipeline # switch to "mps" for apple devices pipe = DiffusionPipeline.from_pretrained("darkstorm2150/Protogen_Infinity_Official_Release", 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
- Local Apps Settings
- Draw Things
- DiffusionBee
Bug in Protogen-Infinity-X8.6 hosted on civit.ai, it generates different images for the same seed!
It's available on civit.ai but it's broken...
@illtellyoulater broken in what way ?
@darkstorm2150 I observed that despite using exactly the same seed and settings, the X8.6 version of Protogen Infinity from civit.ai (b0c941b464) generates different images.
So I started asking other users if they could see the same behavior and someone confirmed it and told me the model was missing the clip component, which I later saw myself by using the model-toolkit extension.
Then I learned that when WebUI tries to load a model lacking a component, it will re-use the same-type component from the model immediately preceding it in terms of chronological order of loading.
This causes that the only way for Protogen to generate the same image for the same seed across multiple generations, is making sure that on future generation attempts, before loading Protogen, we load exactly the same model that was originally loaded before it at the time the original image was generated...
I really don't know how I actually managed to debug this, as initially it made no sense at all and it was incredibly confusing.
I really hope you can fix it asap and save at least the sanity of some other users ;)