import torch
from diffusers import DiffusionPipeline
# switch to "mps" for apple devices
pipe = DiffusionPipeline.from_pretrained("DrHouseFan-315/EFIG-1", dtype=torch.bfloat16, device_map="cuda")
prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k"
image = pipe(prompt).images[0]Ethical Flow Matching Image Generator is a flow-matching based vae and flow model I trained purely on cc0 and public domain images from https://free-images.com/
The goal was to train it on 100% ethical data and have it be recognizeable.
Unfortunately, it still uses clip for text encoding, so it's not fully there.
Combined, the unet and the vae have 102m parameters.
Feel free to use this for whatever, but let me know what you make with it!
The model was trained on my PC, I have solar.
Sample generations, for the prompts a brick house, trans flag, wooden door, grassy field, blue flower, tree, mountain, sunset, pizza
The VAE is pretty good in itself, achieving a 96x compression.
VAE reconstructions
VAE samples
VAE latent interpolation
- Downloads last month
- -