Summary

This chapter tried to solve one of the most discussed topics: overcoming the 77-token limitation and adding prompt weights for the Stable Diffusion pipeline using the Diffusers package. Automatic1111’s Stable Diffusion WebUI provides a versatile UI and is now (as I am writing this) the most prevailing prompt weighting and attention format. However, if we take a look at the code from Automatic1111, we will probably get lost soon; its code is long without clear documentation.

This chapter started with understanding the root cause of the 77-token limitation and advanced to how the Stable Diffusion pipeline uses prompt embeddings. We implemented two functions to overcome the 77-token limitation.

One simple function without weighting was implemented to show how to walk around the 77-token limitation. We also built another function with the full function of a long prompt usage without length limitations and also have prompt weighting implemented.

By understanding...