Sign In
home
models
images
videos
posts
articles
bounties
tools
challenges
events
shop

Another Prompting Guide (for Pony Diffusion)

Raidez's Avatar
Raidez
Dec 2, 2024
generation guide
promptingguidelines
Another Prompting Guide (for Pony Diffusion)

Welcome adventurers, this present missive aims to guide you during your wonderful journey through prompting.
I am still a beginner in this field, but I will still transmit to you a part of my infinite knowledge.

All images in this guide were generated with Pony Diffusion v6 XL (no LoRA, embedding, or anything else), and all image contains ComfyUI workflow, just import it.

Choose your side

Padawan, it's time to choose your side.
Generally, there are two styles of prompts:

  • natural language, you formulate your prompt as a sentence.

  • the other, not really named, but I will call it expectative language, tags/keywords, very close to Danbooru tags.

I choose for you

The rest of this guide will be based on expectative language.

But why ?

Because Pony was trained with booru sites (art libraries), images are tagged like this one.

And the other main reason is that I am lazy, writting in natural language is a pain in the ass (especially for a non-english like me).

So you can look at booru sites as a database of tags, to help you generate images:

  • danbooru : the most famous one.

  • safebooru : a safe version of danbooru, with less NSFW content.

  • just google for "booru"

Another good resource is Danbooru wiki tag groups, it's a good way to find tags by category.

Follow the BASE

My method was largely inspired by the CASE method but with subtle differences.

Don't sue me for copyright taches, I can't afford a lawyer for that 🙃.

It's a simple 4-step process:

  1. Basic : quality, context, camera

  2. Acts : actors, action, posture, gesture

  3. Specs : specs (hair, eyes, ...)

  4. Envs : environment, background, props

Let's look at an example:

score_9, score_8_up, score_7_up, score_6_up,
source_anime, rating_safe,
cowboy shot, from side,

1girl, solo,
sitting, reading,
holding book,

(very dark skin:1.25),
red eye, ginger hair,
embarrassed, sweat,
casual clothes,

indoors, library,
moon, candlelight,
wooden furnitures

Basic

  • quality (score_9, score_8_up, ...): mandatory

  • context

    • source (anime, pony, furry, cartoon)

    • rating (safe, questionable, explicit)

    • artist (when PD7 will come out)

  • camera

    • framing (portrait, cowboy shot, full body)

    • angle (straight-on, from side, from below)

Acts

  • actors (1girl, 1boy, solo, 2girls, solo focus, ...)

  • action (writing, sitting, sleeping, ...): use verb

  • posture, gesture (heads up, arms up, ...)

Specs

  • body (skin, hair, breasts, horns, tails, ...)

  • face (eye, hair)

  • expression (blush, smile, ...)

  • clothes

Envs

  • indoors/outdoors, place (school, bedroom, garage)

  • light, time of day (sunny, dawn, dusk, night)

  • weather (wind, rain, snow)

  • objects (furniture, vehicles, ...)

Extras

Because Civitai don't have table layout, I have tricked with an image, here the images in the "table".

23

Comments