Create Stable Diffusion API on GCP: A Step-by-Step Guide

What is Stable Diffusion?

Stable Diffusion is a cutting-edge deep learning model launched in 2022 that specializes in text-to-image generation. It allows users to create high-quality images from textual descriptions. The model’s flexibility extends to various applications, including inpainting, outpainting, and generating image translations by guiding them with text prompts. With the backing of Stability.ai, Stable Diffusion has quickly grown in popularity among developers and artists alike.

How to Create a GCP Account

To utilize Stable Diffusion effectively, you must create a Google Cloud Platform (GCP) account if you haven’t done so already. Follow these steps:

1. Visit GCP Free Tier and sign up.
2. Set up a billing account since GPU functionalities require billing-enabled accounts.
3. Consider establishing a budget with alerts to avoid unexpected expenses, especially as GPU usage can accumulate rapidly.

Requesting GPU Access

Once your account is set up, follow these steps to request GPU access:

1. Enable the Compute Engine API by navigating to the APIs & Services page and searching for it.
2. Click on the API and then on Enable.
3. In the Quotas section, filter for GPUs (all regions) and request an increase from 0 to 1, explaining that you will be using a Machine Learning model that requires GPU support.
4. Be aware that approval for your request may take up to two days.

Creating a Virtual Machine Instance

After your GPU access is approved, it's time to create a VM instance:

1. Go to GCP Compute Instances and click on Create instance.
2. Name your instance; e.g., "stable-diffusion-instance".
3. Select the region carefully, as GPU availability can vary.
4. Choose a machine type; the NVIDIA T4 is a solid, cost-effective option.
5. For the boot disk, set it to Deep Learning Linux with at least 50GB of storage.
6. Enable the required firewall settings by checking both Allow HTTP traffic and Allow HTTPS traffic.

Creating a Firewall Rule

Before accessing your instance, set up a firewall rule:

1. Go to GCP Firewall Rules and select Create Firewall Rule.
2. Name it stable-diffusion-rule and add your previous tag in the targets section.
3. Set the source IP ranges to 0.0.0.0/0 and allow tcp:5000 through protocols and ports.

Accessing Your Compute Instance

The simplest way to access your newly created VM instance is via SSH from your GCP console. Navigate back to your instance list and click on your instance.

Setting Up Stable Diffusion

Upon your first login, you will be prompted to install the NVIDIA drivers:

1. Follow the prompt by typing ‘Y’ for yes to install drivers.
2. Please note that you'll need to reinstall these drivers after any instance stop/start.
3. Clone the necessary repositories required for Stable Diffusion.
4. Ensure Docker is installed.
5. Finally, build your Docker image and download the model weights from Hugging Face after creating an account and generating an Auth token.

Testing Your Model

Once the installations are complete and the model is running, you will receive an output file (typically output-1.png). You can download it easily via terminal commands.

To find your current working directory, type pwd and note the path for the next steps.

API Testing

Your model's responses will be encoded in base64 format. Use tools like base64 to image converter to decode and preview your images.

Exploring Deforum

If you’re interested in video creation, consider checking out Deforum, a model particularly designed for creating short videos.

Conclusion

Thank you for following this guide on setting up Stable Diffusion on GCP! For more tutorials, feel free to visit our tutorial page and explore further.

These instructions were compiled from various sources to provide comprehensive guidance on achieving your machine learning objectives with Stable Diffusion.