Training Diffusion Models with Reinforcement Learning
Paper • 2305.13301 • Published • 5
import torch
from diffusers import DiffusionPipeline
# switch to "mps" for apple devices
pipe = DiffusionPipeline.from_pretrained("kvablack/ddpo-aesthetic", dtype=torch.bfloat16, device_map="cuda")
prompt = "lion"
image = pipe(prompt).images[0]ddpo-aesthetic
This model was finetuned from Stable Diffusion v1-4 using DDPO and a reward function based on the LAION aesthetic predictor. See the project website for more details.
The model was finetuned for 40 iterations with a batch size of 256 samples per iteration. It was only prompted with the following list of 45 common animals during finetuning, so try those for the best results. However, we also observed limited generalization to other prompts.
Prompts:
- cat
- dog
- horse
- monkey
- rabbit
- zebra
- spider
- bird
- sheep
- deer
- cow
- goat
- lion
- tiger
- bear
- raccoon
- fox
- wolf
- lizard
- beetle
- ant
- butterfly
- fish
- shark
- whale
- dolphin
- squirrel
- mouse
- rat
- snake
- turtle
- frog
- chicken
- duck
- goose
- bee
- pig
- turkey
- fly
- llama
- camel
- bat
- gorilla
- hedgehog
- kangaroo
- Downloads last month
- 2