Model Demo

In this tutorial, we will run Stable Diffusion model to generate an image from text. Specifically, we want the model to generate a photo of an astronaut riding a horse on mars. The image is stored on EverlyAI file storage and it can be downloaded from web UI (opens in a new tab).

Step 1 Implement Code

The final folder structure is shown below.

everlyai_entrypoint.sh is the entrypoint of the server. It contains code to install dependencies and start the server.

python3 -m pip install diffusers transformers accelerate scipy safetensors
python3 stable_diffusion.py

stable_diffusion.py contains the entire image generation code, adapted from Huggingface model card (opens in a new tab).

import torch
from diffusers import StableDiffusionPipeline, EulerDiscreteScheduler
 
model_id = "stabilityai/stable-diffusion-2"
 
# Use the Euler scheduler here instead
scheduler = EulerDiscreteScheduler.from_pretrained(model_id, subfolder="scheduler")
pipe = StableDiffusionPipeline.from_pretrained(model_id, scheduler=scheduler, torch_dtype=torch.float16)
pipe = pipe.to("cuda")
 
prompt = "a photo of an astronaut riding a horse on mars"
image = pipe(prompt).images[0]
    
image.save("/everlyai/fs/astronaut_rides_horse.png")

Inside the project directory, my_awesome_project, run the following command to generate the code.zip file. And now we are ready to start the project on EverlyAI.

zip code.zip everlyai_entrypoint.sh stable_diffusion.py

Step 2 Start Project

Go to EverlyAI project (opens in a new tab) page. Start the project with the following configuration.

Step 3 Verify Result

Wait until the job completes. We can now navigate to file storage (opens in a new tab) page. You should be able to see and download the image from there.