Background
Previously, I failed to get anything running using AMD and ROCm for stable diffusion. One suggestion online was using <a target="_blank" href="https://nod.ai/">nod.ai</a> SHARK, which uses IREE to run ML through Vulkan. SHARK supports Radeon out of the box, so I decided to give it a try. As with most things, I do things the "developer way," so those are the directions I followed. SHARK is a complete solution though, so it's different cobbling and building stuff like llama.cpp or Gaussian Splatting.
<pre><code class="lang-plaintext">git clone https://github.com/nod-ai/SHARK.git
cd .\SHARK\
.\setup_venv.ps1
 cd .\apps\stable_diffusion\web\
python .\index.py
</code></pre>
This, you can point your browser to your <code>localhost:8080</code> and put in prompts like "Harry Potter on a horse"
<img src="https://cdn.hashnode.com/res/hashnode/image/upload/v1695495641671/7146ab1f-4eac-4972-ba9a-5cf95364d287.png" alt class="image--center mx-auto" />
This <code>512x512</code> image took about 20 secs to generate.

Background

Previously, I failed to get anything running using AMD and ROCm for stable diffusion. One suggestion online was using [nod.ai](https://nod.ai/) SHARK, which uses IREE to run ML through Vulkan. SHARK supports Radeon out of the box, so I decided to give it a try. As with most things, I do things the "developer way," so those are the directions I followed. SHARK is a complete solution though, so it's different cobbling and building stuff like llama.cpp or Gaussian Splatting.

```plaintext
git clone https://github.com/nod-ai/SHARK.git
cd .\SHARK\
.\setup_venv.ps1
 cd .\apps\stable_diffusion\web\
python .\index.py
```

This, you can point your browser to your `localhost:8080` and put in prompts like "Harry Potter on a horse"

![](https://cdn.hashnode.com/res/hashnode/image/upload/v1695495641671/7146ab1f-4eac-4972-ba9a-5cf95364d287.png align="center")

This `512x512` image took about 20 secs to generate.

Stable Diffusion, AMD, and nod.ai