luna

This image background generated with stable diffusion luna

Stable diffusion is a deep learning, text-to-image model and used to generate detailted images conditioned on text description, thout it can also be applied to other task such as inpainting or outpainting and generate image to image translate guide by text prompt.

try online on google colab:

luna can running on online colab notebook you can check here:

luna
luna with gpu and mixed precision
- 15 seconds generation time per image (512 x 512) on default colab GPU without drop the quality

usage for online colab:

click the connect
open on runtime and click run all ( ctrl+f9 if using shortcut )
online collab running luna

usage

use venv

//create venv
python3 -m venv venv

//activate venv
source venv/bin/activate

clone repo

clone on https :

git clone https://github.com/slowy07/luna
cd luna
pip install -r requirements.txt

clone on ssh :

git clone git@github.com:slowy07/luna.git
cd luna
pip install -r requirements.txt

note : if using mac m1 you can try installing the requirements_m1.txt

run script

python text2image.py --prompt="example text"

to change the output file name run using the --output flag

python text2image.py --prompt="cool picture" --output="cool_pic.png"

install as python package

pip install git+https://github.com/slowy07/luna

and run the package using

from stable_diffusion_tensorflow.stable_diffusion import Text2Image

generator = Text2Image(img_height=512, img_width=512, jit_compose=False)
img = generator.generate(
  "DSLR photograph of an astronut riding a horse",
  num_steps = 50,
  unconditional_guidance_scale = 75,
  temperature = 1,
  batch_size = 1,
)

you can change dimension of image by change the img_height and img_width

generator = Text2Image(
  img_height = 1020 # or change 1080
  img_height = 1080 # or change 800
)

⚠️ NOTE ⚠️

if pip encountered an issue, try running pip with higher privilege using sudo

output

img = generator.generate(
  "DSLR photograph of an astronut riding a horse",
  num_steps = 50,
  unconditional_guidance_scale = 75,
  temperature = 1,
  batch_size = 1,
)

output:

description	image
prompt: DSLR photograph of an astronut riding a horse
prompt: DSLR photograph of an astronut riding a horse with unconditional_guidance_scale 7.5
prompt: DSLR photograph of an astronut riding a horse with unconditional_guidance_scale 7.5
prompt: DSLR photograph of an astronut riding a horse with unconditional_guidance_scale 7.5

other output

image_prompt: str = "Harry potter playing basketball"
img = generator.generate(
  image_prompt,
  num_steps = 50,
  unconditional_guidance_scale = 5.6,
  temperature = 1,
  batch_size = 1,
)

description	image
prompt: indonesia country with cyberpunk style
prompt: 3d detailed woman knight
prompt: 3d detailed woman knight (next generation process)
prompt: 3d detailed woman knight (next generation process)
prompt: 3d detailed woman knight (next generation process)
prompt: realistic detailed knight
prompt: aesthetic art haunted house
prompt: medival castle realistic and detailed oil painting
prompt: harry potter playing basketball
prompt: harry potter playing guitar
prompt: person hacking with computer
prompt: God wearing mask
prompt: Harry potter random
prompt: Marilyn monroe with random art style

rafiqola/luna