wuerstchen-prior-model-base

<img src="https://cdn-uploads.huggingface.co/production/uploads/634cb5eefb80cc6bcaf63c3e/i-DYpDHw8Pwiy7QBKZVR5.jpeg" width=1500>

Würstchen - Overview

Würstchen is a diffusion model, whose text-conditional model works in a highly compressed latent space of images. Why is this important? Compressing data can reduce computational costs for both training and inference by magnitudes. Training on 1024x1024 images is way more expensive than training on 32x32. Usually, other works make use of a relatively small compression, in the range of 4x - 8x spatial compression. Würstchen takes this to an extreme. Through its novel design, we achieve a 42x spatial compression. This was unseen before because common methods fail to faithfully reconstruct detailed images after 16x spatial compression. Würstchen employs a two-stage compression, what we call Stage A and Stage B. Stage A is a VQGAN, and Stage B is a Diffusion Autoencoder (more details can be found in the paper). A third model, Stage C, is learned in that highly compressed latent space. This training requires fractions of the compute used for current top-performing models, allowing also cheaper and faster inference.

Würstchen - Prior

The Prior is what we refer to as "Stage C". It is the text-conditional model, operating in the small latent space that Stage A and Stage B encode images into. During inference, its job is to generate the image latents given text. These image latents are then sent to Stages A & B to decode the latents into pixel space.

Prior - Model - Base

This is the base checkpoint for the Prior (Stage C). This means this is only pretrained and generates mostly standard images. We recommend using the interpolated model, as this is our best checkpoint for the Prior (Stage C) because it was finetuned on a curated dataset. However, we recommend this checkpoint if you want to finetune Würstchen on your own large dataset, as the other checkpoints are already biased towards being more artistic. This checkpoint should provide a fairly standard baseline to finetune from, as long as your dataset is rather large.

Note: This checkpoint was also already trained on multi-aspect-ratios, meaning you can generate larger images than just 1024x1024. Sometimes generations up to 2048x2048 even work. Feel free to try it out!

Also Note: The base checkpoint usually requires a higher classifier-free-guidance value (guidance_scale=8.0) and also a negative caption in order to make good looking images. The interpolated model and finetuned model usually don't need a negative caption and work better with a lower classifier-free-guidance value (guidance_scale=4.0).

Image Sizes

Würstchen was trained on image resolutions between 1024x1024 & 1536x1536. We sometimes also observe good outputs at resolutions like 1024x2048. Feel free to try it out. We also observed that the Prior (Stage C) adapts extremely fast to new resolutions. So finetuning it at 2048x2048 should be computationally cheap. <img src="https://cdn-uploads.huggingface.co/production/uploads/634cb5eefb80cc6bcaf63c3e/IfVsUDcP15OY-5wyLYKnQ.jpeg" width=1000>

How to run

This pipeline should be run together with https://huggingface.co/warp-ai/wuerstchen:

import torch
from diffusers import WuerstchenDecoderPipeline, WuerstchenPriorPipeline
from diffusers.pipelines.wuerstchen import WuerstchenPrior, default_stage_c_timesteps

device = "cuda"
dtype = torch.float16
num_images_per_prompt = 2

prior = WuerstchenPrior.from_pretrained("warp-ai/wuerstchen-prior-model-base", torch_dtype=dtype).to(device)
prior_pipeline = WuerstchenPriorPipeline.from_pretrained(
    "warp-ai/wuerstchen-prior", prior=prior, torch_dtype=dtype
).to(device)
decoder_pipeline = WuerstchenDecoderPipeline.from_pretrained(
    "warp-ai/wuerstchen", torch_dtype=dtype
).to(device)

caption = "Anthropomorphic cat dressed as a fire fighter"
negative_prompt = "bad anatomy, blurry, fuzzy, extra arms, extra fingers, poorly drawn hands, disfigured, tiling, deformed, mutated, drawing"

prior_output = prior_pipeline(
    prompt=caption,
    height=1024,
    width=1024,
    timesteps=default_stage_c_timesteps,
    negative_prompt=negative_prompt,
	guidance_scale=8.0,
    num_images_per_prompt=num_images_per_prompt,
)
decoder_output = decoder_pipeline(
    image_embeddings=prior_output.image_embeddings,
    prompt=caption,
    negative_prompt=negative_prompt,
    num_images_per_prompt=num_images_per_prompt,
    guidance_scale=0.0,
    output_type="pil",
).images

Model Details

• Developed by: Pablo Pernias, Dominic Rampas

• Model type: Diffusion-based text-to-image generation model

• Language(s): English

• License: MIT

• Model Description: This is a model that can be used to generate and modify images based on text prompts. It is a Diffusion model in the style of Stage C from the Würstchen paper that uses a fixed, pretrained text encoder (CLIP ViT-bigG/14).

• Resources for more information: GitHub Repository, Paper.

• Cite as:

  @misc{pernias2023wuerstchen,
        title={Wuerstchen: Efficient Pretraining of Text-to-Image Models}, 
        author={Pablo Pernias and Dominic Rampas and Marc Aubreville},
        year={2023},
        eprint={2306.00637},
        archivePrefix={arXiv},
        primaryClass={cs.CV}
  }
  

Environmental Impact

Würstchen v2 Estimated Emissions Based on that information, we estimate the following CO2 emissions using the Machine Learning Impact calculator presented in Lacoste et al. (2019). The hardware, runtime, cloud provider, and compute region were utilized to estimate the carbon impact.

• Hardware Type: A100 PCIe 40GB
• Hours used: 24602
• Cloud Provider: AWS
• Compute Region: US-east
• Carbon Emitted (Power consumption x Time x Carbon produced based on location of power grid): 2275.68 kg CO2 eq.