Image Generation with Python and Google Colab

Introduction to Image Generation with Python

Image generation is a fascinating field in artificial intelligence, where models are trained to create new images from scratch. This can be done using various techniques, such as Generative Adversarial Networks (GANs), Variational Autoencoders (VAEs), or diffusion models. In this guide, we will use a pre-trained model called Stable Diffusion to generate images from text prompts.

Prerequisites

Before we start, ensure you have the following:

1. Google Account: You need a Google account to use Google Colab.
2. Basic Python Knowledge: Familiarity with Python programming.
3. Understanding of Deep Learning: Basic knowledge of neural networks and deep learning concepts.
4. Google Colab: A cloud-based Jupyter notebook environment provided by Google.

Step-by-Step Guide to Image Generation in Google Colab

Step 1: Open Google Colab

1. Go to Google Colab.
2. Click on “New Notebook” to create a new Colab notebook.

Step 2: Set Up the Environment

1. Enable GPU: To speed up the image generation process, enable the GPU.
   • Go to Runtime > Change runtime type.
   • Select GPU under the “Hardware accelerator” dropdown.
   • Click Save.
2. Install Required Libraries:

   !pip install diffusers transformers torch accelerate

   This installs the diffusers library (for Stable Diffusion), transformers (for text processing), torch (PyTorch), and accelerate (for optimization).

Step 3: Import Libraries

In the next cell, import the required libraries:

import torch
from diffusers import StableDiffusionPipeline
from PIL import Image

Step 4: Load the Pre-trained Stable Diffusion Model

Load the pre-trained Stable Diffusion model. We’ll use the StableDiffusionPipeline from the diffusers library.

# Load the Stable Diffusion model
pipe = StableDiffusionPipeline.from_pretrained("runwayml/stable-diffusion-v1-5", torch_dtype=torch.float16)
pipe = pipe.to("cuda")

• torch_dtype=torch.float16 ensures the model uses half-precision floating points, which is faster and consumes less memory.
• pipe.to("cuda") moves the model to the GPU for faster computation.

Step 5: Generate an Image from a Text Prompt

Now, let’s generate an image using a text prompt. For example, let’s generate an image of “a futuristic cityscape at sunset.”

# Define the prompt
prompt = "a futuristic cityscape at sunset"

# Generate the image
with torch.autocast("cuda"):
    image = pipe(prompt).images[0]

# Display the image
image.show()

• torch.autocast("cuda") ensures mixed precision is used for faster computation.
• pipe(prompt).images[0] generates the image and retrieves the first (and only) image from the output.

Step 6: Save the Generated Image

To save the generated image, use the following code:

# Save the image
image.save("generated_image.png")

This saves the image as generated_image.png in your current working directory.

Step 7: Experiment with Different Prompts

You can experiment with different text prompts to generate various images. For example:

prompt = "a serene mountain landscape with a lake"
image = pipe(prompt).images[0]
image.show()

Step 8: (Optional) Customize Image Generation Parameters

You can customize the image generation process by adjusting parameters like num_inference_steps, guidance_scale, etc. For example:

image = pipe(prompt, num_inference_steps=50, guidance_scale=7.5).images[0]
image.show()

• num_inference_steps: Controls the number of denoising steps (higher values produce better quality but take longer).
• guidance_scale: Controls how closely the image follows the prompt (higher values make the image more aligned with the prompt).

Step 9: Share or Download the Image

You can download the generated image from Google Colab:

1. Click on the folder icon on the left sidebar to open the file explorer.
2. Locate the generated_image.png file.
3. Click the three dots next to the file and select “Download.”

Conclusion

You have successfully created an image generation pipeline using Stable Diffusion in Google Colab! This is just the beginning—you can explore more advanced models, fine-tune them on custom datasets, or integrate them into larger applications.

Full Code for Reference

# Step 1: Install required libraries
!pip install diffusers transformers torch accelerate

# Step 2: Import libraries
import torch
from diffusers import StableDiffusionPipeline
from PIL import Image

# Step 3: Load the pre-trained Stable Diffusion model
pipe = StableDiffusionPipeline.from_pretrained("runwayml/stable-diffusion-v1-5", torch_dtype=torch.float16)
pipe = pipe.to("cuda")

# Step 4: Generate an image from a text prompt
prompt = "a futuristic cityscape at sunset"
with torch.autocast("cuda"):
    image = pipe(prompt).images[0]

# Step 5: Display and save the image
image.show()
image.save("generated_image.png")

Feel free to experiment and have fun generating images!