Character LoRAs - Want more character variety? Then check This List!
Description

It may look like a virtual reality helmet, but this point remains for your imagination: perhaps a high-tech element has been added to a simple harness, or maybe it’s just a simple cover(lid) that completely blocks vision while performing certain tasks, like a falconry hood.
Yes, in fact, this is not quite the correct result for the query, but I just liked it aesthetically.
Prompt: feral, mlp, pony, bridle, blinders, harness, ponyplay, horse blinders, eyes, bridle with blinders
Negative: boring_e621_fluffyrock_v4, deformityv6, easynegative, bad-hands-5, bad anatomy, low quality, signature, logo, watermark, human, sleeping mask, blindfold
Size: 1536x1280 (upscaled from 768x640)
Sampler: DPM++ 2M Karras (20 steps) CFG Scale: 5.00 Denoising Strength: 0.75
Model: molKeunKemoXL Seed: 168372860

suggestive4948 ai generated41572 generator:stable diffusion21414 prompter:mechanic3127 oc4733 oc only4071 earth pony7744 pony20454 ambiguous gender48 blindfold106 bridle173 cutie mark144 ears up136 halter24 harness143 orange hair49 orange mane26 orange tail7 pony play16 prompt in description1024 reins9 sensory deprivation4 solo31372 submissive1233 tack213 yellow body14 yellow coat40 yellow skin59
Source

Comments

Syntax quick reference: **bold** *italic* ||hide text|| `code` __underline__ ~~strike~~ ^sup^ ~sub~

Detailed syntax guide

@Mechanic31
Yeah, I originally was copying and pasting big negative prompts, but started picking them apart and realised a lot of things in them didn’t work or if they did, only worked because any text added changes the picture slightly. Some times they were even making the picture worse. (One reason I tend to stick to a fixed seed is so that I can see the exact effects changing the prompt is having easily.)
Even when they do work, they only work on models trained with those particular keywords, too.
These days, I tend to just start with “signature, watermark”, and add things as needed. Though illustrious models do want some of the negative tags like “low quality”, so it really does vary by model…
(human, sleeping mask, and blindfold all sound like things that probably actually were helpful in the negative for this picture, though.)
Several of the keywords you’re using in the negative are for triggering embeddings for 1.5 based models, and given XL is in the model name, they probably aren’t doing much, since 1.5 embeddings don’t work on xl and vice-versa. Any text affects the final image, but you could likely safely remove “boring_e621_fluffyrock_v4, deformityv6, easynegative, bad-hands-5,” from the negative, as all of those are embeddings. And that pony doesn’t have hands anyways, bad or otherwise…