NoobAI-XL (NAI-XL)

Name: NoobAI-XL (NAI-XL)
Rating: 0 (0 reviews)
Author: L_A_X

2.9k

34.0k

87.3k

671

Updated: Nov 3, 2024

base model

anime based styles

Images hidden due to mature content settings

Download (6.46 GB)

Verified: 20 days ago

SafeTensor

Details

Type	Checkpoint Trained
Stats	1,583
Reviews	Very Positive (158)
Published	Oct 16, 2024
Base Model	SDXL 1.0
Hash	AutoV2 D4E7A4249C

1 File

About this version

NoobAI XL (V-pred branch)
NoobAI XL (V预测分支)

This model page is the V-pred branch of NoobAI XL, trained with the EA version followed by the 8 epoch version, which cannot be used in AUTOMATIC1111 WebUI. Please use it via diffusers or reForge.

This test was mainly conducted by @Euge_, thanks to his hard work as a member of Laxhar Lab.

该模型页面为 NoobAI XL 的 V 预测分支，使用Early Access Ver加训8ep的版本训练而成，无法在 AUTOMATIC1111 WebUI 中使用。请通过 diffusers 或 reForge 使用。本测试由@尤吉主要进行，感谢尤吉作为Laxhar Lab成员的辛勤付出ミ(・・)ミ

Usage: reForge

Install and launch reForge, and choose branch;
git checkout dev_upstream_experimental
Find “Advanced Model Sampling for Forge” at the bottom of the page;
Enable “Enable Advanced Model Sampling”;
Select “v_prediction” in “Discrete Sampling Type”.

用法：reForge

安装并启动 reForge,并使用命令切换分支；
git checkout dev_upstream_experimental
在页面下方找到 “Advanced Model Sampling for Forge”；
启用 “Enable Advanced Model Sampling”；
在 “Discrete Sampling Type” 中选择 “v_prediction”。

Usage: Diffusers
用法：Diffusers

import torch

from diffusers import StableDiffusionXLPipeline, EulerAncestralDiscreteScheduler

ckpt_path = "/path/to/model.safetensors"

pipe = StableDiffusionXLPipeline.from_single_file(

ckpt_path,

use_safetensors=True,

torch_dtype=torch.float16,

)

pipe.scheduler = EulerAncestralDiscreteScheduler.from_config(pipe.scheduler.config)

pipe.scheduler.register_to_config(

prediction_type="v_prediction",

rescale_betas_zero_snr=True,

)

pipe.enable_xformers_memory_efficient_attention()

pipe = pipe.to("cuda")

prompt = "best quality, 1boy, solo"

negative_prompt = "bad hands, worst quality, low quality, bad quality, multiple views, 4koma, comic, jpeg artifacts, monochrome, sepia, greyscale, flat color, pale color, muted color, low contrast, bad anatomy, picture frame, english text, signature, watermark, logo, patreon username, web address, artist name"

image = pipe(

prompt=prompt,

negative_prompt=negative_prompt,

width=832,

height=1216,

num_inference_steps=28,

guidance_scale=7.0,

generator=torch.Generator().manual_seed(42),

).images[0]

image.save('image.png')

L_A_X

License:

CreativeML Open RAIL++-M Addendum

Images hidden due to mature content settings

this is an image generation model based on training from Illustrious-xl, and continued trained by Laxhar Lab.

https://civitai.com/models/795765/illustrious-xl

It utilizes the latest full Danbooru and e621 datasets for training, with native tags caption.

The version uploaded on 8 October trained 5 epochs on 8*H100, as a Early Access Version.

And huggingface page of Lab

New Version is training on 32*H100

https://huggingface.co/Laxhar/sdxl_noob

Follow-up models and technical reports will be posted on huggingface

Communication qq groups：①875042008 ②914818692 ③635772191

DISCORD: Laxhar Dream Lab SDXL NOOB

We extend our sincere gratitude to our esteemed GPU sponsors for their generous support: https://cloud.lanyun.net

This version of the model improves the fit of the characters and styles in Illustrious-xl 0.1ver, and the specific characteristics of the characters have a better representation. Laxhar lab is currently continuing to train the new version of the open-source model of XL on the basis of this beta version in the hope of minimising the use of lora, and releasing a more Noob-friendly, one-click SDXL anime model!

Note: The model name and other details are subject to change.

This model is still undergoing training!!!

Current Status

This is a 100% EPS version released for use. Our next steps include developing a v-prediction version and a dedicated controlNet for NoobAI-XL, as well as additional complementary plugins, to enhance the model's versatility. This is also the original intention of this model—to be 'a model that even noobs can use well.

Datasets

- Danbooru (Pid: 1~7,600,039):

https://huggingface.co/datasets/KBlueLeaf/danbooru2023-webp-4Mpixel

- Danbooru (Pid > 7,600,039):

https://huggingface.co/datasets/deepghs/danbooru_newest-webp-4Mpixel

- E621 Data as of 2024-04-07 :

https://huggingface.co/datasets/NebulaeWis/e621-2024-webp-4Mpixel

Caption

<1girl/1boy/1other/...>, <character>, <series>, <artists>, <special tags>, <general tags>

Quality Tags

For quality tags, we evaluated image popularity through the following process:

Data normalization based on various sources and ratings.
Application of time-based decay coefficients according to date recency.
Ranking of images within the entire dataset based on this processing.

Our ultimate goal is to ensure that quality tags effectively track user preferences in recent years.

Percentile Range Quality Tags

> 95th masterpiece

> 85th, <= 95th best quality

> 60th, <= 85th good quality

> 30th, <= 60th normal quality

<= 30th worst quality

In the CCIP test, noobaiXL showed an improvement of approximately 2% compared to its base model. Based on data from over 3,500 characters, 89.2% of the characters achieved a CCIP score higher than 0.9. Given the current model performance, it is necessary to further expand the dataset for the existing CCIP test.

NoobAI-XL Short Test Report V0.1

https://nx9nemngdhk.feishu.cn/docx/XcAddUhDOo57U7x7MbXcE6VNnYc

Monetization Prohibition:

● You are prohibited from monetizing any close-sourced fine-tuned / merged model, which disallows the public from accessing the model's source code / weights and its usages.

● As per the license, you must openly publish any derivative models and variants. This model is intended for open-source use, and all derivatives must follow the same principles.