ⓘ

Yiffy-e18 Style Tag Comparison

This webpage displays pre-generated images from an AI text-to-picture generator, showing how a single tag can affect otherwise identical images.

These images were created with the Stable Diffusion UI and the Yiffy-e18 model. You can use this software and the below settings to generate these results for yourself!

Example 1: The Bathing Bear

Prompt: [input tag], explict, chubby, furry, anthro, bear, female, large breasts, long wavy hair, river, wet, wading, [input tag]
Negative Prompt: None
Initial Image: None
Seed: 1029050
Model: yiffy-e18
Sampler: plms
Image Size: 512 x 512
Inference Steps: 128
Guidance Scale: 8.0
Fix incorrect faces and eyes: No
Upscale Image: RealESRGAN_x4plus

Example 2: The Office Corgi

Prompt: [input tag], explict, furry, anthro, corgi, male, penis, balls, solo, nude, office, hair, [input tag]
Negative Prompt: None
Initial Image: None
Seed: 2468
Model: yiffy-e18
Sampler: plms
Image Size: 512 x 512
Inference Steps: 128
Guidance Scale: 6.0
Fix incorrect faces and eyes: No
Upscale Image: RealESRGAN_x4plus

Prompt Theory

You may notice that "explicit" is misspelled in both prompts. This is on purpose, since the Yiffy-e18 engine was trained with a typo. This is also based on the rules of the e621-style rating system (safe/questionable/explicit).

The [input tag] in the prompt represents the variable tag in the prompt, and is included both at the start and at the end. The reason for this is twofold: First, it adds emphasis, and that seems to be stronger than using brackets for emphasis. Second, word order matters, and it seems to me that the prompt is parsed in a procedural fashion. Tags at the start of prompts seem to apply to the underlying structure of the image (body shape, composition), and tags at the end of prompts apply more to the finishing touches (brushstrokes, shading and such). Putting these tags at both ends of the prompt makes the image start and end with the intended style.

Credits

Layout and interface by NaniMoose
Special thanks to Cobalt for testing, suggestions, and general help
All images were produced by Artificial Intelligence, and trample all rights everywhere into a fine powder