Update: Added optimized_img2img.py

This code can now generate a 512x512 image from prior images using less than 4GB of RAM (in less than 20 seconds on RTX 2060)

Optimized Stable Diffusion (Sort of)

This repo is a modified version of the Stable Diffusion repo, optimized to use lesser VRAM than the original by sacrificing on inference speed.
It has two python files. 1) optimized_img2img.py to generate new image based on a given image and prompt. 2) optimized_txt2img.py to generate an image based only on a prompt.

img2img

It can generate 512x512 images from a prior image and prompt on a 4GB VRAM GPU in under 20 seconds per image (RTX 2060 in my case).
You can use the --H & --W arguments to set the size of the generated image.The maximum size that can fit on 6GB GPU (RTX 2060) is around 576x768.
For example, the following command will generate 20 512x512 images:

python optimizedSD/optimized_img2img.py --prompt "Austrian alps" --init-img ~/sketch-mountains-input.jpg --strength 0.8 --n_iter 2 --n_samples 5 --H 576 --W 768

txt2img

It can generate 512x512 images from a prompt on a 4GB VRAM GPU in under 25 seconds per image (RTX 2060 in my case).
You can use the --H & --W arguments to set the size of the generated image.
For example, the following command will generate 20 512x512 images:

python optimizedSD/optimized_txt2img.py --prompt "Cyberpunk style image of a Telsa car reflection in rain" --H 512 --W 512 --seed 27 --n_iter 2 --n_samples 10 --ddim_steps 50

To get the lowest inference time per image, use the maximum batch size --n_samples that can fit on the GPU. Inference time per image will reduce on increasing the batch size, but the required VRAM will also increase.
If you get a CUDA out of memory error, try reducing the batch size --n_samples. If it doesn't work, the other option is to reduce the image width --W or height --H or both.
Mixed Precision is enabled by default. If you don't have a GPU with tensor cores, you can still use mixed precision to run the code using lesser VRAM but the inference time may be larger. And if you feel that the inference is slower, try using the --precision full argument to disable it.
All the modified files are in the optimizedSD folder, so if you have already cloned the original repo, you can just download and copy this folder into the orignal repo instead of cloning the entire repo.
You can also clone this repo and follow the same installation steps as the original written below (mainly creating the conda env and placing the weights at the specified location).

To achieve this, the stable diffusion model is fragmented into four parts which are sent to the GPU only when needed. After the calculation is done, they are moved back to the CPU. This allows us to run a bigger model on a lower VRAM.
The only drawback is higher inference time which is still an order of magnitude faster than inference on CPU.

Stable Diffusion

Stable Diffusion was made possible thanks to a collaboration with Stability AI and Runway and builds upon our previous work:

High-Resolution Image Synthesis with Latent Diffusion Models
Robin Rombach*, Andreas Blattmann*, Dominik Lorenz, Patrick Esser, Björn Ommer

which is available on GitHub.

Stable Diffusion is a latent text-to-image diffusion model. Thanks to a generous compute donation from Stability AI and support from LAION, we were able to train a Latent Diffusion Model on 512x512 images from a subset of the LAION-5B database. Similar to Google's Imagen, this model uses a frozen CLIP ViT-L/14 text encoder to condition the model on text prompts. With its 860M UNet and 123M text encoder, the model is relatively lightweight and runs on a GPU with at least 10GB VRAM. See this section below and the model card.

Requirements

A suitable conda environment named ldm can be created and activated with:

conda env create -f environment.yaml
conda activate ldm

You can also update an existing latent diffusion environment by running

conda install pytorch torchvision -c pytorch
pip install transformers==4.19.2
pip install -e .

Stable Diffusion v1

Stable Diffusion v1 refers to a specific configuration of the model architecture that uses a downsampling-factor 8 autoencoder with an 860M UNet and CLIP ViT-L/14 text encoder for the diffusion model. The model was pretrained on 256x256 images and then finetuned on 512x512 images.

*Note: Stable Diffusion v1 is a general text-to-image diffusion model and therefore mirrors biases and (mis-)conceptions that are present in its training data. Details on the training procedure and data, as well as the intended use of the model can be found in the corresponding model card. Research into the safe deployment of general text-to-image models is an ongoing effort. To prevent misuse and harm, we currently provide access to the checkpoints only for academic research purposes upon request. This is an experiment in safe and community-driven publication of a capable and general text-to-image model. We are working on a public release with a more permissive license that also incorporates ethical considerations.*

Request access to Stable Diffusion v1 checkpoints for academic research

Weights

We currently provide three checkpoints, sd-v1-1.ckpt, sd-v1-2.ckpt and sd-v1-3.ckpt, which were trained as follows,

sd-v1-1.ckpt: 237k steps at resolution 256x256 on laion2B-en. 194k steps at resolution 512x512 on laion-high-resolution (170M examples from LAION-5B with resolution >= 1024x1024).
sd-v1-2.ckpt: Resumed from sd-v1-1.ckpt. 515k steps at resolution 512x512 on "laion-improved-aesthetics" (a subset of laion2B-en, filtered to images with an original size >= 512x512, estimated aesthetics score > 5.0, and an estimated watermark probability < 0.5. The watermark estimate is from the LAION-5B metadata, the aesthetics score is estimated using an improved aesthetics estimator).
sd-v1-3.ckpt: Resumed from sd-v1-2.ckpt. 195k steps at resolution 512x512 on "laion-improved-aesthetics" and 10% dropping of the text-conditioning to improve classifier-free guidance sampling.

Evaluations with different classifier-free guidance scales (1.5, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0) and 50 PLMS sampling steps show the relative improvements of the checkpoints:

Text-to-Image with Stable Diffusion

Stable Diffusion is a latent diffusion model conditioned on the (non-pooled) text embeddings of a CLIP ViT-L/14 text encoder.

After obtaining the weights, link them

mkdir -p models/ldm/stable-diffusion-v1/
ln -s <path/to/model.ckpt> models/ldm/stable-diffusion-v1/model.ckpt

and sample with

python scripts/txt2img.py --prompt "a photograph of an astronaut riding a horse" --plms

By default, this uses a guidance scale of --scale 7.5, Katherine Crowson's implementation of the PLMS sampler, and renders images of size 512x512 (which it was trained on) in 50 steps. All supported arguments are listed below (type python scripts/txt2img.py --help).

usage: txt2img.py [-h] [--prompt [PROMPT]] [--outdir [OUTDIR]] [--skip_grid] [--skip_save] [--ddim_steps DDIM_STEPS] [--plms] [--laion400m] [--fixed_code] [--ddim_eta DDIM_ETA] [--n_iter N_ITER] [--H H] [--W W] [--C C] [--f F] [--n_samples N_SAMPLES] [--n_rows N_ROWS]
                  [--scale SCALE] [--from-file FROM_FILE] [--config CONFIG] [--ckpt CKPT] [--seed SEED] [--precision {full,autocast}]

optional arguments:
  -h, --help            show this help message and exit
  --prompt [PROMPT]     the prompt to render
  --outdir [OUTDIR]     dir to write results to
  --skip_grid           do not save a grid, only individual samples. Helpful when evaluating lots of samples
  --skip_save           do not save individual samples. For speed measurements.
  --ddim_steps DDIM_STEPS
                        number of ddim sampling steps
  --plms                use plms sampling
  --laion400m           uses the LAION400M model
  --fixed_code          if enabled, uses the same starting code across samples
  --ddim_eta DDIM_ETA   ddim eta (eta=0.0 corresponds to deterministic sampling
  --n_iter N_ITER       sample this often
  --H H                 image height, in pixel space
  --W W                 image width, in pixel space
  --C C                 latent channels
  --f F                 downsampling factor
  --n_samples N_SAMPLES
                        how many samples to produce for each given prompt. A.k.a. batch size
  --n_rows N_ROWS       rows in the grid (default: n_samples)
  --scale SCALE         unconditional guidance scale: eps = eps(x, empty) + scale * (eps(x, cond) - eps(x, empty))
  --from-file FROM_FILE
                        if specified, load prompts from this file
  --config CONFIG       path to config which constructs model
  --ckpt CKPT           path to checkpoint of model
  --seed SEED           the seed (for reproducible sampling)
  --precision {full,autocast}
                        evaluate at this precision

Note: The inference config for all v1 versions is designed to be used with EMA-only checkpoints. For this reason use_ema=False is set in the configuration, otherwise the code will try to switch from non-EMA to EMA weights. If you want to examine the effect of EMA vs no EMA, we provide "full" checkpoints which contain both types of weights. For these, use_ema=False will load and use the non-EMA weights.

Image Modification with Stable Diffusion

By using a diffusion-denoising mechanism as first proposed by SDEdit, the model can be used for different tasks such as text-guided image-to-image translation and upscaling. Similar to the txt2img sampling script, we provide a script to perform image modification with Stable Diffusion.

The following describes an example where a rough sketch made in Pinta is converted into a detailed artwork.

python scripts/img2img.py --prompt "A fantasy landscape, trending on artstation" --init-img <path-to-img.jpg> --strength 0.8

Here, strength is a value between 0.0 and 1.0, that controls the amount of noise that is added to the input image. Values that approach 1.0 allow for lots of variations but will also produce images that are not semantically consistent with the input. See the following example.

Input

Outputs

This procedure can, for example, also be used to upscale samples from the base model.

Comments

Our codebase for the diffusion models builds heavily on OpenAI's ADM codebase and https://github.com/lucidrains/denoising-diffusion-pytorch. Thanks for open-sourcing!
The implementation of the transformer encoder is from x-transformers by lucidrains.

BibTeX

@misc{rombach2021highresolution,
      title={High-Resolution Image Synthesis with Latent Diffusion Models},
      author={Robin Rombach and Andreas Blattmann and Dominik Lorenz and Patrick Esser and Björn Ommer},
      year={2021},
      eprint={2112.10752},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}

jcaw / stable-diffusion