Hello everyone, I hope you have a nice day. :)
I would like to share my personal project. A few months back, I started a mini project to showcase the range of models and loras on my Twitter. It has evolved into quite a big passion project with many upgrades done towards the template, I've also started to post on Pinterest to be able to have a cleaner overview for all showcase. Now I've decided to also post on Civitai under the models itself so the showcase could be viewed directly, if the post is lost in the sea of images, all of the post can still be found on my post page.
I hope that my little project will not just be my interest but I am able to help everyone compare the results and find your favorite models/loras.
General Guidelines
How do I decide which model and version to test?
It's all random, if you would like me to showcase your model, you can let me know. Sometimes I will only post one version of a multi-version model, most likely it was an old model that I downloaded prior, I will download and test other version when I have the time.
How do I decide what to prompt for a model test?
I keep the prompt as minimal as possible and do not use lora so the model can generate the image as close as to their base style. Categories are decided to test the range of the outcome, such as can they generate good animals? An overview categories are gender comparison, clip comparison, ratio comparison, color range, scenery range, objects and living creatures.
How do I decide which result to use?
Aside from those with fixed seeds(character), I'll test 4 result with random seeds before upsizing the one that I believe produce the best result, if all 4 results are not up to standard, I'll do another 4 and choose between all 8 of them, if they are still not up to standard, I'll choose the one closest to the prompt.
Do I use any extension?
I use ADetailer on the face if there is a character in the result, I think looks are important when deciding which model to use. I do not fix hands as some models can naturally generate better hands, so I want to showcase those model's capability.
What does the "NSFW" stamp mean?
Some showcase models have the "NSFW" stamp, that means NSFW(nudity) can be generated without being prompted. The "NSFW" stamp does not mean that the models are able/unable to generate NSFW, more solely to keep creators aware that NSFW generations might be occur even when unprompted in case they would want to use it in a public setting.
Basic Setting
---Generation Setting---
Sampling Steps: 30
Sampling Method: DPM++ 2M Karras
CFG scale: 7
SD VAE: kl-f8-anime2.ckpt
---Hires. fix---
Denoising strength: 0.5
Hires upscale: 2
Hires steps: 15
Hires upscaler: 4x-UltraSharpLora Showcase
The purpose of the lora showcase is to see how well lora works on different type of model and how would the weight affect it.
Prompts
The prompt is just ((best quality, masterpiece, absurbres)),the lora and the keyword, nothing else! The goal is to see how does well does the lora work by itself. (I might make changes to this.)
Seed
A range of seed will be first tested, estimated between 10~20 across various models in different weights first, from there I'll pick the one which I feel produce the best result.
Model Type
The model will be categorized into a few category, art style, character, concept, fashion, and others. Others will include anything that could not fit into the prior category, such as sliders, looks, expression, filter etc.
Weight Range
6 range of weight will be generated, including the control, usually 0.0, the maximum recommended weight, usually 1.0, the remaining will be spaced out equality in between.
In cases of a slider lora for a bigger range, e.g -1.0 weight to 1.0 weight, the maximum recommended weight, 1.0 and the minimum recommended weight, -1.0, the remaining will be spaced out equality in between.
Type of Models
Anime Model - The style of the model is similar to anime/manga style.
2.5 CGI Model - The style of the model creates semi-realistic style. (I might make changes to this.)
3D CGI Model - The style of the model creates 3D rendering style.
Realistic Model - The style of the model creates photorealistic style, easy to confuse with actual photo.
Model Showcase
Version 2.0 onwards
Improvements:
-to be edited-
Character Prompt (CLIP 1 & CLIP 2)
This category is the control for all models. The all settings, prompts and seeds will be fixed. It will be easy to see the difference of all variety of models.
Prompt: ((best quality, masterpiece, absurbres, super-resolution)) beautiful, movement, alluring. smiling, starry sky, ambience, moon, night sky, colorful lighting, cinematic lighting, neon light glowing, medium shot
Opposite Gender Prompt*: ((best quality, masterpiece, absurbres, super-resolution)) Boy, handsome, movement, alluring. smiling, starry sky, ambience, moon, night sky, colorful lighting, cinematic lighting, neon light glowing, medium shot
Seed(Fixed): 3231210346
Resolution: 512, 768 [2:3]
*If the first generation generate a male character, the opposite gender prompt will be reversed from boy to girl, and handsome to pretty.Other Character Prompt
Chibi (Clip 1) - Chibi prompt will be emphasized upwards from Chibi to (Chibi), (Chibi:1), (Chibi:1.3), (Chibi:1.5), (Chibi:1.8), (Chibi:2.0), respectively until a chibi character us generated.
Prompt:((best quality, masterpiece, absurbres, super-resolution)) Chibi, beautiful, movement, alluring. smiling, starry sky, ambience, moon, night sky, colorful lighting, cinematic lighting, neon light glowing, medium shot
Seed: Random
Resolution: 512, 768 [2:3]Monster (Clip 1) - This category test if any type of monster can be generated by the model, it will could range from cute devil girls to animal beasts to small creatures to giant monsters. If after the maximum attempts and there are still no result, the image that closest symbolize a monster will be chosen.
Prompt:((best quality, masterpiece, absurbres, super-resolution)) Monster
Seed: Random
Resolution: 512, 768 [2:3]Colors Prompt (CLIP 1 & CLIP 2)
This category not only test the potential color range, but also try to show if the model heavily weights towards any certain colors. It also helps to explore the model's biases to generate character/background/layout in both Clip 1 and Clip 2.
Prompt: ((best quality, masterpiece, absurdres, super-resolution)) (_______:1.5)
CLIP 1 Variation: Warm Colors, Cool Colors, Neutral Colors
CLIP 2 Variation: Pastel Colors, Colorful, Dark Colors
Seed: Random
Resolution: 512, 512 [1:1]-to be edited-
Other Prompt
This category test random subjectsin both organic and geometric shape. It also test the generation ability of the model to see if they excel in any aspect. If after the maximum attempts and there are still no result, the image that closest symbolize a monster will be chosen.
Food(Clip 1) Prompt: ((best quality, masterpiece, absurbres, super-resolution)) Food photography, table full of food
Vehicle(Clip 1) Prompt: ((best quality, masterpiece, absurbres, super-resolution)) Vehicle
Seed: Random
Resolution: 512, 512 [1:1]Mecha(Clip 1) Prompt: ((best quality, masterpiece, absurdres, super-resolution)) Mecha
Animal(Clip 1) Prompt: ((best quality, masterpiece, absurdres, super-resolution)) Animal
Seed: Random
Resolution: 768, 432 [16:9]Scenery Character Prompt (CLIP 1 & CLIP 2)
This category showcases the ability to generate sceneries, in both portrait and landscape view.
The sceneries are decided are in hopes to showcase the models ability to create between nature vs man-made scenery and day vs night scenery. It is also able shows if the model will prioritize character over background. If after the maximum attempts and there are still no result, the image that closest symbolize a monster will be chosen.
Fun fact: Volcano, Rainforest and Underwater was choosen base of the four base element, fire, earth and water. Don't worry, I did not forget about air, but my interpretation is more of the skyline, day and night.
Volcano (Clip 1) Prompt: ((best quality, masterpiece, absurbres, super-resolution)), Detailed Scenery, Disaster, Volcano, Lava
Rainforest (Clip 1) Prompt: ((best quality, masterpiece, absurbres, super-resolution)), Detailed scenery, Rainforest
Underwater (Clip 1) Prompt: ((best quality, masterpiece, absurbres, super-resolution)), detailed scenery, coral reef, underwater , cinematic landscape
Bedroom (Clip 2) Prompt: ((best quality, masterpiece, absurbres, super-resolution)), Interior Design, Aesthetic Bedroom
Asian Traditional City (Clip 2) Prompt: ((best quality, masterpiece, absurbres, super-resolution)), Detailed historical background, Asian traditional city
Galactic (Clip 2) Prompt: ((best quality, masterpiece, absurbres, super-resolution)), Detailed galactic scenery, outer space, faraway planets
Seed: Random
Resolution: 768, 512 [3:2]Night Cityscape (Clip 1) Prompt: ((best quality, masterpiece, absurbres, super-resolution)) detailed scenery, buildings, nightlife, aerial view
Mountain Skyline (Clip 2) Prompt: ((best quality, masterpiece, absurbres, super-resolution)) detailed scenery, mountains, skyline, clouds
Seed: Random
Resolution: 432, 768 [9:16]Version 0.11 to 1.24 (old version)
Character Prompt (CLIP 1 & CLIP 2)
This category is the control for all models. The all settings, prompts and seeds will be fixed. It will be easy to see the difference of all variety of models.
Prompt: ((best quality, masterpiece, absurbres)) beautiful, movement, alluring. smiling, starry sky, ambience, moon, night sky, colorful lighting, cinematic lighting, neon light glowing, medium shot
Seed(Fixed): 3231210346
Resolution: 512, 768 [2:3]Chibi (CLIP 1)
This category will try to generate chibi, sometimes when chibi are hard to achieve, emphasis will be increased to (((Chibi))) or ((Chibi:1)). This category also will test image generation for 1:1 ratio aspect.
How I rate chibify-able?
YES - If the chibi is within a 1 to 3 head ratio and there is no uncanny resemblance. The subject should also does not have a dismembered body.
MAYBE - Able to generate "YES" with high emphasis on 25% of the result.
NO - Unable to generate "YES" even with high emphasis.
Prompt: ((best quality, masterpiece, absurbres)), (colorful palette:1.5), ((Chibi)), movement, alluring. smiling, starry sky, ambience, moon, night sky, neon light glowing, ((surrounded by stuffed animal))
Seed(Fixed): 1394183393
Resolution: 768, 768 [1:1]Scenery (CLIP 1)
This category showcases the ability to generate sceneries, in both portrait and landscape view on a 9:16 ratio aspect. The test is to show if the scenery would be simple or detailed.
It is also able shows if the model prioritize character, for those models, characters will always be visible in the scenery, either as a main subject or as part of the scenery.
Landscape(Horizontal) Prompt: ((best quality, masterpiece, absurbres)), amazing landscape, detailed scenery, aesthetic background, starry sky, shooting stars, big moon, night sky, colorful lighting, cinematic lighting, neon light glowing, depth of field
Portrait(Vertical) Prompt: ((best quality, masterpiece, absurbres)), amazing landscape, detailed scenery, aesthetic background, depth of field
Seed: Random
Resolution: 768, 432 [16:9] / 432, 768 [9:16]Animal - Cat-able? (CLIP 1)
This category showcase the model's ability to generate a cat using it's style, it also showcases the ability to get the correct anatomy of an animal. Why cat? After trying out different prompts, cats are the easiest to prompt throughout models.
How I rate cat-able?
YES - If the cat does not have a dismembered body and is the main subject focus.
MAYBE - Able to generate "YES" with 50% chance of being the main subject.
NO - Unable to generate "YES", will only generate cat girls or girls with cats.
Prompt: ((best quality, masterpiece, absurbres, super-resolution)) cute illustration (((animal, cute cat)))
Seed: Random
Resolution: 512, 768 [2:3]Color Palette (R,O,Y,G Clip 1, T,B,P,P Clip2)
This category not only test the potential color range, but also try to show if the model heavily weights towards any certain colors. It also helps to explore the model's base ability to generate character/background/layout in both Clip 1 and Clip 2.
Prompt: ((best quality, masterpiece, absurdres, super-resolution)) (______ colors:1.1)
CLIP 1 Variation: Red, Orange, Yellow, Green
CLIP 2 Variation: Teal, Blue, Purple, Pink
Seed: Random
Resolution: 512, 768 [2:3]Tone Color Palette (W,C,N CLIP 1)
This category is similar to the color palette but it shows more of the tone of the result. It can show if the model is more heavily weighted to generate warmer colors, cooler colors or neutral. Example some models general neutral in a more beige(warm) tone, while others in a more bluish gray(cool) tone.
Prompt: ((best quality, masterpiece, absurdres, super-resolution)) (______ colors:1.1)
Variation: Warm, Cool, Neutral
Seed: Random
Resolution: 768, 512 [3:2]Version 1 Changelogs
30/08/23 - v1.24
1. Change the prompt of Character(CLIP 1), Character(CLIP 2), Chibi, Vertical Landscape & Horizontal Landscape from "((best quality, masterpiece, absurbres))" to "((best quality, masterpiece, absurbres, super-resolution))" for all.
2. Change the prompt of Color Tones from "((best quality, masterpiece, absurdres, super-resolution)) (______ colors:1.2)" to "((best quality, masterpiece, absurdres, super-resolution)) (______ colors:1.1)"
3. Change the prompt of Color Palette from "((best quality, masterpiece, absurdres, super-resolution)) (_____ color palette:1.1) to "((best quality, masterpiece, absurdres, super-resolution)) (______ colors:1.1)"
29/08/23 - v1.23
1. Changed the title from "Cat-able?" to "No Human, Cat-able?"
18/08/23 - v1.22
1. Change the title from "Color" to "Color Theme".
2. Added Color Key layer.
20/08/23 - v1.21
1. Changed the prompt of Horizontal Landscape from "((best quality, masterpiece, absurbres)), amazing landscape, detailed scenery, aesthetic background, starry sky, shooting stars, big moon, night sky, colorful lighting, cinematic lighting, neon light glowing, gradient Purple green sky, depth of field" to "((best quality, masterpiece, absurbres)), amazing landscape, detailed scenery, aesthetic background, starry sky, shooting stars, big moon, night sky, colorful lighting, cinematic lighting, neon light glowing, depth of field"
2. Changed the prompt of Color Tone form "((best quality, masterpiece, absurdres, super-resolution)) (______ Tone color:1.4)" to "((best quality, masterpiece, absurdres, super-resolution)) (______ colors:1.2)"
3. Include Ratio Aspect for images.
08/08/23 - v1.11
1. Changed the prompt of Vertical Landscape from "((best quality, masterpiece, absurbres)), amazing landscape, detailed scenery, aesthetic background, starry sky, shooting stars, big moon, night sky, colorful lighting, cinematic lighting, neon light glowing, gradient Purple green sky, depth of field" to "((best quality, masterpiece, absurbres)), amazing landscape, detailed scenery, aesthetic background, depth of field"
05/08/23 - v1.10
1. Removed Seed name for Landscape, Chibi, Cats, Tones, and Colors.
2. Renamed Landscapes to Vertical Landscape and Horizontal Landscape.
3. Updated Title from Pink/Purple to Blue/Purple
4. Updated Twitter Logo to X
22/07/23 - v1.00
1. Increased dimension from 3100x3500px to 3950x3550px.
2. Added - Vertical Landscape , Warm Tone, Cool Tone and Neutral Tone examples.
3. Color Palette, Changed Monochrome to Teal.
4. Updated Color Palette Red, Orange, Yellow Green with Clip 1
5. Updated Color Palette Teal, Blue, Purple, Pink with Clip 2
6. Updated Logo.
7. Updated Title with water effect.
20/07/23 - v0.12
1. Changed 3126x3504px dimension to 3100x3500px.
19/07/23 - v0.11
1.Changed Title to a lighter purple.This article's purpose is meant to help keep me in check with all the experiments and expansion of this project.