afiaka87 / StyleGANXL-CLIP

A notebook for text-based guided image generation using StyleGANXL and CLIP.

StyleGANXL+CLIP notebook

A notebook for Out-of-Domain text-based image generation using the new StyleGAN-XL¹ and CLIP².

This is a work in progress. Currently, the main improvement points are:

• Find a way to stabilize and improve the generation process, since it tends to become noisy and chaotic.
• Figure out the best optimizer and it's values.

Examples

Two judo athletes fighting	An ant	A red apple

Acknowledgements

The notebook is largely based on code by Katherine Crowson and nshepperd.

We are also very grateful for the help of the CasualGANPapers community and especially Axel Sauer for taking it's time to help us and answer our questions.

Footnotes

1. StyleGANXL was created by Axel Sauer and Katja Schwarz and Andreas Geiger. Here is the official implementation. ↩

2. CLIP (Contrastive Language-Image Pre-Training) is a multimodal model made by OpenAI. For more information head over here. y ↩