Text-to-Image AI based models are all the hype at the moment. With many commercial organisations dabbling into this space like Bing, Adobe, Google etc. You also have dedicated services provided by sites like Midjourney (the leader in my view), Leonardo.ai, DreamStudio etc that provide paid subscription to create images using their platform.
However, you are here because you want to experiment in house, create your images locally, train your own models and run batch processing as long as you like. Even though I’ve been using some of these services and also renting/subscribing to Cloud GPU, there are many limitations with such services. They might have time limitations, credit allowance, GPU availability and limited availability to install extensions.
Specifications of Custom PC
The ideal use case is then to run your OWN custom PC that will be able to run AI model to generate Text-to-Image. I have place the order for a custom built PC with the following specifications.
- AMD Ryzen 7 5700X 8 core processor
- Asus TUF Gaming B550-PLUS Wifi II ATX Motherboard
- 32 GB of Kingston Fury Beast DDR4 RAM
- Gigabyte RTX 4080 16GB Gaming OC GFXcard
- CM MasterLiquid ML240L ARGB V2 CPU Cooler
- Kensigton 1TB NV2 PCIe Gen4 SSD
- CM MWE 850W Power Supply
- Windows 11 Pro OEM
The team as JW Computers Castle Hill will be building this for me and I should have it in the next 2 weeks. Overall cost of this build is AU$ 3,600 of which $1,940 is the cost of the nvidia GPU.
My goals are to be able to run the following.
- Run Stable Diffusion using Automatic1111
- Run custom models of Stable Diffusion
- Train my own models
- Run Google Colabs to play with StyleGAN
- Run Deforum Text2Video Automatic1111 Extension
- Run and test Lora, ControlNet etc
If you are not following me you can follow me on Twitter and I will be posting more updates on this blog.