Enabling long prompts with weighting
We just built a whatever size of text encoder for a Stable Diffusion pipeline (v1.5-based). All of those steps are paving the way to build long prompts with a weighting text encoder.
A weighted Stable Diffusion prompt refers to the practice of assigning different levels of importance to specific words or phrases within a text prompt used for generating images through the Stable Diffusion algorithm. By adjusting these weights, we can control the degree to which certain concepts influence the generated output, allowing for greater customization and refinement of the resulting images.
The process typically involves scaling up or down the text embedding vectors associated with each concept in the prompt. For instance, if you want the Stable Diffusion model to emphasize a particular subject while deemphasizing another, you would increase the weight of the former and decrease the weight of the latter. Weighted prompts enable us to better direct...
