Type | |
Stats | 535 |
Reviews | (90) |
Published | Mar 30, 2024 |
Base Model | |
Hash | AutoV2 369D6D9D05 |
Welcome to my RE:COLLECTION.
I am exploring every possibility I can do in this passionate hobby of mine. I am still slow to learn, if there is anything I am not doing enough, I would love to get your help and feedback (I am not very IT savvy). If you want to share any of my model on other sites, I am happy to let you do it as long as it is not for monetization, but please inform me as I may already have account on some sites.
In this sixth addition to my series, "RE:LAXATION" was merged to fill the hole for a style that is between 2D and 2.5D in my collection.
It is a gentle model that generates 2.3D/2.5D characters, they are simple and smooth, a departure from my previous abstract style. This model loves to create soft colors and pastel colors on default, it can also get a little naughty sometimes if your prompts get too dangerous. This model also prefers to keep generations clean with minimal noise unless otherwise prompt, so if you are looking for detailed background or intricate design, you will need to emphasis your prompt for it. As my usual, please inpaint whenever needed to fix freaky generations.
I do not do NSFW test with my model, if they appear unintentionally, just add "NSFW" in the negative prompt.
Switch between version to compare the difference.
Image comparison range: Image 2 to Image 13
VERSION:
SIDEA VERSION - The main iteration of RE:LAXATION model.
SIDE25A VERSION - Not related to SIDEA, but a more semi-realistic RE:LAXATION version.
SIDEB VERSION - Sub-iteration of RE:LAXATION model.
SIDE25B VERSION - Calmer semi-realistic version, compared to SIDE25A.
EXTRA VERSION - Extra-iteration, for those who still prefers a more anime style.
[NEW] EXTRA 25 VERSION - A slightly different art style, feels more cool than gentle.
RE:COLLECTION Model Series:
*From the 5th model(RE:IMGAGINATION) onwards, I've not test the LITE model, result between the full and pruned version may slightly vary. All showcase are generated with the full version.
Please try around with your settings, but if you need a guide, this is my settings:
---Generation Setting---
Sampling Steps: 30
Sampling Method: DPM++ 2M Karras
CFG scale: 7
Clip skip: 1 or 2
SD VAE: kl-f8-anime2.ckpt
---Hires. fix---
Denoising strength: 0.5
Hires upscale: 2 ~ 2.5
Hires steps: 15
Hires upscaler: 4x-UltraSharp
There isn't any fixed prompt, but I can recommend to try these:
Positive Prompt: best quality, masterpiece, absurbres, super-resolution
Negative Prompt: badhandv4, easynegative, ng_deepnegative_v1_75t, verybadimagenegative_v1.3, (worst quality, low quality:1.4),text, words, logo, watermark
Disclaimer: I do not bake my models with VAEs or embedding as some will prefer to use their own, if you want my result, just follow my setting and prompts.