prior-weightupdate

Finetuning - daehan17/prior-weightupdate

This pipeline was finetuned from kandinsky-community/kandinsky-2-2-prior on the lambdalabs/naruto-blip-captions dataset. Below are some example images generated with the finetuned pipeline using the following prompts: ['A robot pokemon, 4k photo']:

Pipeline usage

You can use the pipeline like so:

from diffusers import DiffusionPipeline
import torch

pipe_prior = DiffusionPipeline.from_pretrained("daehan17/prior-weightupdate", torch_dtype=torch.float16)
pipe_t2i = DiffusionPipeline.from_pretrained("kandinsky-community/kandinsky-2-2-decoder", torch_dtype=torch.float16)
prompt = "A robot pokemon, 4k photo"
image_embeds, negative_image_embeds = pipe_prior(prompt, guidance_scale=1.0).to_tuple()
image = pipe_t2i(image_embeds=image_embeds, negative_image_embeds=negative_image_embeds).images[0]
image.save("my_image.png")

Training info

These are the key hyperparameters used during training:

• Epochs: 7
• Learning rate: 1e-05
• Batch size: 1
• Gradient accumulation steps: 4
• Image resolution: 768
• Mixed-precision: None

More information on all the CLI arguments and the environment are available on your wandb run page.