diffuse-my-lyrics 0.0.2

A dummy project to generate images from song lyrics using Latent Stable Diffusion

Diffuse My Lyrics!

🎶 ➡ 🧠 ➡ 🖼️

An easy way to generate images from lyrics

Description

This is a simple application that uses the spectacular Stable Diffusion model to generate images from song lyrics. So just install the library on a Colab notebook, choose your favorite song and sit back and wait for the visual interpretations of each verse!

In the Cool Outputs section (only in repository README.md), I have shown some interpretations of verses that I found very cool 😏😏😏

How To Use

First, simply install the python package in a colab notebook (right now, it's only for Colab, but extending it to general use is trivial ... as long as you have a good GPU 😅 ).

# Install the latest version of the package
$ pip install -U diffuse-my-lyrics

Now, suppose we want to feed the model with the following verses, belonging to The End, a magnificent piece by The Doors.

Ride the King's highway, baby
Weird scenes inside the gold mine
Ride the highway west, baby
Ride the snake, ride the snake
To the lake, the ancient lake, baby
The snake is long, seven miles
Ride the snake, he's old, and his skin is cold

After uploading this lyrics to the colab notebook (I am using a .txt extension), we just need to run the following commands.

# Import the Lyrics2Images class
from diffuse_my_lyrics import Lyrics2Images

l2i = Lyrics2Images(num_inference_steps=100) # In this case, we are indicating the model to run for 100 steps
l2i.run(input_path="/content/my_favourite_song.txt", output_path="my_favourite_song_folder")

After running Lyrics2Images, a folder will be created in your colab current directory (my_favourite_song_folder), where a series of images will be generated (one image for each verse of the lyrics).

One it's finished, simply zip the folder and download it!!

import shutil
shutil.make_archive("zipped_folder", 'zip', "my_favourite_song_folder")

Arguments

• model_id - The model id. By default CompVis/stable-diffusion-v1-4
• revision - The model revision. By default fp16
• torch_dtype - The Pytorch dtype. By default torch.float16
• prompt - This parameter is useful if you want to add additional information to the verse. For example, digital art, HQ, etc. By default digital art
• num_inference_steps - The number of steps. By default 50
• use_auth_token - This parameter determines whether to use an authentication token for Hugging Face. By default True

Next Steps

• Add support for generating several images instead of just one.
• Make the library usable in another environments (not just Colab)
• Create argument for using a manual seed
• Add custom size of output images