Sign In

Double Exposure SDXL

33
141
18
11
Verified:
SafeTensor
Type
LoRA
Stats
141
18
Reviews
Published
May 8, 2024
Base Model
SDXL 1.0
Usage Tips
Clip Skip: 1
Strength: 1
Hash
AutoV2
A5CD52C741

Experimental LoRA to see if controllable double exposure (superimposition) generations are possible via simple natural language captions.

Results:

Yes, it is possible to generate superimposed images by describing the different image 'layers' of the full image.

However, The LoRA would benefit greatly from a larger dataset.


Prompting tips:

There is no prompt template. Generally, describe the full image you want with natural language. Although, you can use the comma-delimited (SD1.5) prompting method as well. In this case, ensure you include double exposure or superimposed somewhere in the prompt. Not using natural language sacrifies some control.


Limitations:

  • Prompt Adherence: Due to insufficient training data; the generated image may not appear exactly as you described.

  • As a LoRA, the quality and prompt adherence are going to be influenced by the base model you're using. In other words, the same prompt will be different depending on which base model you're using. I found that some of the purely photorealistic models struggle with the abstract concept of a multi-exposure image.


Questions/Feedback/Updates?

Visit my thread on the Unstable Diffusion Discord


Links:

Find all epochs at 🤗Huggingface.

Checkout my other models

SDXL

LoRA SDXL

SD1.5

LoRA SD1.5