Overcoming the 77-token limitation using community pipelines

Implementing a pipeline supporting long prompt weighting from scratch can be challenging. Often, we simply wish to utilize Diffusers to generate images using detailed and nuanced prompts. Fortunately, the open source community has provided implementations for SD v1.5 and SDXL. The SDXL implementation was originally initialized by Andrew Zhu, the author of this book, and massively improved by the community.

I’ll now provide two examples of how to use the community pipeline for SD v1.5 and SDXL:

1. This example uses the lpw_stable_diffusion pipeline for SD v1.5.

   Use the following code to start a long prompt weighted pipeline:

   from diffusers import DiffusionPipeline

   import torch

   model_id_or_path = "stablediffusionapi/deliberate-v2"

   pipe = DiffusionPipeline.from_pretrained(

       model_id_or_path,

       torch_dtype = torch.float16,

       custom_pipeline =...