IronJayx/serverless-pix2pix

🍌 Serverless pix2pix

Ready to deploy inference endpoint for pix2pix.

Disclaimer

This repo uses the HuggingFace diffusers' implementation of Tim Brooks et al. Instruct Pix2pix model - https://www.timothybrooks.com/instruct-pix2pix

How to interact with the service

Run the server

2 options:

Without docker

pip install -r requirements.txt
python3 server.py

Or with docker

docker build -t pix2pix .
docker run -p 8000:8000 pix2pix

Model Inputs

The model accepts the following inputs:

• prompt (str, required)
• image (base64 str, required) - A base64 string of the image (data:image/type;base64,.... also accepeted) should be 512x512 or another standard Stable Diffusion 1.5 resolution for best results
• seed (int, optional, defaults to 42)
• text_cfg_scale (float, optional, default 7)
• image_cfg_scale (float, optional, default 1.5)
• steps (int, optional, default to 20)
• randomize_cfg (boolean, optional, default False)
• randomize_seed (boolean, optional, default True)
• image_cfg_scale (float, optional, default 1.5)

Additional parameters:

• test_mode (boolean, optional, default False)
• toDataUrl (boolean, optional, default False) - if you want output "data:image/type;base64,...."

Not implemented

• negative_prompt
• num_images_per_prompt

Model Output

The model outputs:

A list of image objects where each has the following properties:

• image (base64 str) - base64 or base64 with data_url prefix if specified
• seed (int)
• text_cfg_scale (float)
• image_cfg_scale (float)
• steps (int)

Example

• Checkout the test.py for an example

Examples of generated images

Venus de Milo	Turn her into a cyborg

Elon	Turn him into a cyborg

Helpful Links

Learn more about Instruct Pix2Pix here - https://www.timothybrooks.com/instruct-pix2pix

And Hugging Face support there - https://huggingface.co/timbrooks/instruct-pix2pix

Understand the 🍌 Serverless framework and functionality of each file within it.

Use Banana for scale.