AI Image Generator

Generate AI images on M1/M2 macs in only a few seconds.

Prompt: A realistic photo of a german shepard riding a motorcycle through Tokyo at night

Usage

To run

# Run this once before running the image generator script
source venv/bin/activate

# Generate an image with interactive prompt
python generate.py

# OR
# Run the script directly with arguments
python main.py \
  "a realistic photo of a german shepard riding a motorcycle through Tokyo at night" \
  --steps 10 --width 512 --height 512

# Setting the --continuous flag will cause the script to keep outputting images
# until cancelled by the user

Hugging Face Hub model files are cached in ~/.cache/huggingface

There's over 5GB of data.

du -sh ~/.cache/huggingface

Run latent consistency models on your Mac

Latent consistency models (LCMs) are based on Stable Diffusion, but they can generate images much faster, needing only 4 to 8 steps for a good image (compared to 25 to 50 steps). Simian Luo et al released the first Stable Diffusion distilled model. It’s distilled from the Dreamshaper fine-tune by incorporating classifier-free guidance into the model’s input.

You can run Latent Consistency Models in the cloud on Replicate, but it's also possible to run it locally.

Prerequisites

You’ll need:

a Mac with an M1 or M2 chip
16GB RAM or more
macOS 13.0 or higher
Python 3.10 or above

Install

Run this to clone the repo:

git clone https://github.com/replicate/latent-consistency-model.git
cd latent-consistency-model

Set up a virtualenv to install the dependencies:

python3 -m pip install virtualenv
python3 -m virtualenv venv

Activate the virtualenv:

source venv/bin/activate

(You'll need to run this command again any time you want to run the script.)

Then, install the dependencies:

pip install -r requirements.txt

Run

The script will automatically download the SimianLuo/LCM_Dreamshaper_v7 (3.44 GB) and safety checker (1.22 GB) models from HuggingFace.

python main.py \
  --prompt "a beautiful apple floating in outer space, like a planet" \
  --steps 4 --width 512 --height 512

You’ll see an output like this:

Output image saved to: output/out-prompt-a_beautiful_apple_floating_in_outer_space,_like_a_planet-time-20231027-181911-seed-7445-width-512-height-512-steps-4.png
Using seed: 48404
100%|███████████████████████████| 4/4 [00:00<00:00,  5.54it/s]

Run With A Prompt File

You can also run the script with a prompt file.

Create a prompt file with a few creative prompts:

echo "a beautiful apple floating in outer space, like a planet" > prompts.txt
echo "a knight in armor, with a sword and shield made of jelly" >> prompts.txt
echo "a cat with a human face" >> prompts.txt

Then run the script with the --prompt-file option:

python main.py \
  --prompt-file prompts.txt \
  --steps 4 --width 512 --height 512

You'll see output like this:

Using seed: 58067
100%|████████████████████████████████████| 4/4 [00:01<00:00,  3.28it/s]
Output image saved to: output/out-prompt-A_steampunk_city_at_sunset-time-20231027-181942-seed-58067-width-512-height-512-steps-4.png
Using seed: 13995
100%|████████████████████████████████████| 4/4 [00:00<00:00,  5.19it/s]
Output image saved to: output/out-prompt-A_dragon_playing_chess_with_a_knight-time-20231027-181943-seed-13995-width-512-height-512-steps-4.png

Options

Parameter	Type	Default	Description
--prompt	str	N/A	A text string for image generation.
--prompt-file	str	N/A	A text file with one prompt per-line.
--width	int	512	The width of the generated image.
--height	int	512	The height of the generated image.
--steps	int	8	The number of inference steps.
--seed	int	None	Seed for random number generation.
--continuous	flag	False	Enable continuous generation.

mattdesl / latent-consistency-model