DreamLand (Base)-2.8D-Series-PAseer

QQ group: 816567536 come and have comfultable chat ~~

【ILL2-PATNROM_ILL2_V7】

PATNROM is a half_merged Illustrious 2.0 based CKPT which trained on L20 for about 120h and designed for high-precision prompts generation.

First, It excels in gender distinction, with the balance gender datasets, the clip pollution is reduced to ensure the clear differentiation among male, female, and others. Therefore the problem of [when you use 1boy or 1other or an animal or an robot, there will always a female or sissy feminine character appear] has been solved.

Then, The model delivers exceptional detail in anime or illustration and thick-painting, also closely mimicking professional artists' styles. And maintaining dynamic poses and natural language adaptability.

Finally, this CKPT is a Illustrious 2.0 based CKPT, so ILL2.0 based LORAs and ILL0.1 based LORAs are all repeatable and flexible in this CKPT.

~~Meanwhile, with strong style sensitivity, it replicates distinctive Danbooru aesthetics (only need your know the Secret Words-tags).

The suggestion of parameters :

CFG: [ Dynamic CFG tool ] from 5 to 8

Steps: With my Accelerator LORA : 16 steps. Without my Accelerator LORA: 25 steps.

Prompts Version : 80% tags + 20% natural language.

It takes time and costing to run a GPU card and train the CKPT. So if you like my CKPT you can buy me a cup of coffee or help defencing my electricity bill at https://ko-fi.com/asseer

…………………………………………………………………………………………………………………………………………………

【ILL-PAsaROM V2】

Reduce the over high contrast of the images problem in V1.

Added some NoobAI's databases ignore the images which score are lower than 90 repeats.

The images prompts are generated by PAseer-PromptsGenerater or PAseer-PromptsTransTool. You can use them free on huggingface-space online.

【ILL】

Base on Illustrious Animation checkpoint, Trained on over 20K images for more than 100h on L20 GPU (about $360 on machine).

If you like this checkpoint, please give a like and welcom to post some of your imaginations.

For unknown reason or because base checkpoint, there always be underfitting noises, so use SD Upscaler to fit the noises.

And also suggest to use PAseer-ILL-Accelerater to speed up generating.

If you have no idea how to use this model on ComfyUI, you can use my workflow without SD upscaler. OR workflow with SD upscaler.

This series aming to keep balance between 2D animation and 2.5D render.

Reducing the problem that sometime to generate a female under a male prompt.

If you want to support my contribute, welcom to donate on Kofi to give me a cup of coffe.

【PONY】

Base on PonyV6, AAAAutism | Pony Finetune mix, EDG From Zero

And trained, based on about 12G images, for 20h on L20 GPU.

Also, this is not a idiosyncratic checkpoint, all I hope is flexible for my LORA to generate 2D-2.8D style.

Less necessary for 【score_...】 prompts, but if you added it, there should be nothing happen.

Not able to generate 【ink】 things, I am trying to solve this problem.