Skip to content

Instantly share code, notes, and snippets.

@seefood

seefood/cheatsheet-SDmodels.md

Last active June 28, 2024 17:23
Show Gist options
  • Save seefood/3fdf2a2c2be0b9fa191f2fca2aca0cc9 to your computer and use it in GitHub Desktop.
Save seefood/3fdf2a2c2be0b9fa191f2fca2aca0cc9 to your computer and use it in GitHub Desktop.
Download ZIP
Cheatsheet for Stable Diffusion models
Raw
cheatsheet-SDmodels.md

Cheatsheet for Stable Diffusion models

Common negative prompts

out of frame, duplicate, watermark, signature, text, ugly, morbid, mutated, deformed, blurry, bad anatomy, bad proportions, cloned face, disfigured, fused fingers, fused limbs, too many fingers, long neck,

Models

Pick one in the models dropdown and see special instructions:

NextPhoto

Go over Usage guide, lots of important stuff: https://civitai.com/models/84335/nextphoto

Double Exposure

dublex style or just dublex

Arcane Diffusion

arcane style

Robo Diffusion

nousr robot towards the beginning of your prompt to invoke the finetuned style. Use negative prompts to achieve the best result.

Openjourney

mdjrny-v4 style

funko-diffusion

funko style

Anything v3

1girl, white hair, golden eyes, beautiful eyes, detail, flower meadow, cumulonimbus clouds, lighting, detailed sky, garden. also see https://huggingface.co/Linaqruf/anything-v3.0

redshift-diffusion

redshift style

mo-di-diffusion

modern disney style

nitrosocke/classic-anim-diffusion

classic disney style

DGSpitzer/Cyberpunk-Anime-Diffusion

After model loaded, use keyword dgs in your prompt, with illustration style to get even better results. More: https://huggingface.co/DGSpitzer/Cyberpunk-Anime-Diffusion

Van Gogh Diffusion

This is a fine-tuned Stable Diffusion model (based on v1.5) trained on screenshots from the film Loving Vincent. Use the token lvngvncnt at the BEGINNING of your prompts to use the style (e.g., "lvngvncnt, beautiful woman at sunset"). This model works best with the Euler sampler (NOT Euler_a).

Nitro Diffusion

a fine-tuned Stable Diffusion model trained on three artstyles simultaniously while keeping each style separate from the others. This allows for high control of mixing, weighting and single style use. Use the tokens archer style, arcane style or modern disney style in your prompts for the effect.

GeneralAwareness/MirrorMask

mirrormask

dreamlike.art

Use the same prompts as you would for SD 1.5. Add dreamlikeart if the artstyle is too weak. Non-square aspect ratios work better for some prompts. If you want a portrait photo, try using a 2:3 or a 9:16 aspect ratio. If you want a landscape photo, try using a 3:2 or a 16:9 aspect ratio. Use slightly higher resolution for better results: 640x640px, 512x768px, 768x512px, etc. Also see https://dreamlike.art/

VikingPunk

Similar to the Egyptian styled one, this one is more focused on cooler environments and viking+cyberpunk themes. Works fine for space environments as well, like Alien. Use keyword: VikingPunk

ProGamerGov/knollingcase-embeddings-sd-v2-0

The knollingcase style is considered to be a concept inside a sleek (sometimes scifi) display case with transparent walls, and a minimalistic background.

Suggested prompts:

[concept], micro-details, photorealism, photorealistic, [kc-vx-iter] photorealistic [concept], very detailed, scifi case, [kc-vx-iter] [concept], very detailed, scifi transparent case, [kc-vx-iter]

Suggested negative prompts:

blurry, toy, cartoon, animated, underwater, photoshop

Suggested samplers:

DPM++ SDE Karras (used for the example images) or DPM++ 2S a Karras

Comic Diffusion

From: https://huggingface.co/ogkalu/Comic-Diffusion

Suggested tokens

The tokens for V2 are:

  • charliebo artstyle
  • holliemengert artstyle
  • marioalberti artstyle
  • pepelarraz artstyle
  • andreasrocha artstyle
  • jamesdaly artstyle

LoRA Models

Gustav Dore

https://huggingface.co/lora-library/gustave-dore-dantes-inferno

try prompts like [some subject] as depicted in dantes_inferno and use a v1.5 model

To DL: DL the large .pkl file into models/Lora/dantes_inferno.safetensors

add_detail

standard lora syntax, add <lora:add_detail:NUM> to the prompt, where the NUM is usually 1 but can go negative too.

LowRA

Low key compositions. https://civitai.com/models/48139/lowra

Your start point (weight) is 0.6 ➜ <lora:LowRA:0.6>

Art Deco design

https://huggingface.co/lora-library/artdecodsgn

Add to prompt ➜ in artdecodsgn style, and select a v2.1 model!

Soft-Ones

alvdansen/soft-ones

(and other LoRas by alvdansen), triggered by "daiton" or "daiton style" but not always needed

Embeddings (Textual inversion)

Use these in the prompt, but be aware they are specific to either v1.5 or v2.0 models, so some work with older-based models and some only with the new.

Negative prompt embeddings for SD v 1.5 models

bad-artist-anime bad-artist bad_prompt_version2

use something like sketch by bad-artist or photo by bad_prompt_version2 in the negative prompt input.

Positive embeddings for SD v 1.5 models

madhubani-art see demo: https://huggingface.co/sd-concepts-library/madhubani-art

Positive embeddings for SD v 2.x models

I hope the names are decriptive enough. let me know if you want me to elaborate (If I can find where I got them from...)

v2-0-ActionHelper
v2-0-CGI_Animation-185
v2-0-CGI_Animation-215
v2-0-CGI_Animation-245
v2-0-CinemaHelper
v2-0-KittyPic-1000cp
v2-0-KittyPic
v2-0-PhotoHelper
v2-0-VikingPunk
v2-0-VintageHelper-1000
v2-0-VintageHelper-600
v2-0-anthro
v2-0-kc32-v4-5000
v2-0-kc8-v4-5000
v2-0-mdjrny-papercut
v2-0-midjourney
v2-0-remix

More models and embeddings

Dig here:

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment