Making your own specific AI text2image model
I’ve been playing with LLM-based generative models a lot since last year, and I think they’re really cool !
I really like Dreambooth, but I wanted to see how it compares to other techniques. So I've conducted a comprehensive benchmark comparing:
- DreamBooth,
- Hypernetworks,
- LoRa,
- Textual Inversion,
- Naive fine-tuning
for Stable Diffusion models. This is especially relevant to generative AI and AI Art hobbyists & professionals,
You can find the benchmark and results in this shared GSheet:
I'll try to keep it up to date and make it as accurate and complete as possible.
So, please share it with your colleagues and feel free to provide any feedback or suggestions for improvement 🙂
This is an album cover in Pink Floyd’s style I generated using a fine-tuned Stable Diffusion Model (DreamBooth). Isn’t it cool ?