Text-to-image finetuning - sayakpaul/new_sd_pokemon
This pipeline was finetuned from CompVis/stable-diffusion-v1-4 on the lambdalabs/pokemon-blip-captions dataset. Below are some examples images that were generated with the finetuned pipeline (with the following prompts: ['cute dragon creature', 'cute pokemon creature', 'blue pokemon']):
Pipeline usage
You can use the pipeline like so:
from diffusers import DiffusionPipeline
import torch
pipeline = DiffusionPipeline.from_pretrained("sayakpaul/new_sd_pokemon", torch_dtype=torch.float16)
prompt = "cute dragon creature"
image = pipeline(prompt).images[0]
image.save("my_image.png")
Training info
Following are the key hyperparameters that were used to run finetuning:
- Epochs: 1
- Learning rate: 1e-05
- Batch size: 1
- Gradient accumulation steps: 4
- Image resolution: 512
- Mixed-precision: fp16
More information on all the CLI arguments and the environment should be available on the wandb
run page if you used it via report_to="wandb"
.
Check it out here: https://wandb.ai/sayakpaul/text2image-fine-tune/runs/rb2l1p9o.