NoobAI-XL (NAI-XL)
NoobAI-XL generates high-quality images using Danbooru and e621 datasets, requiring specific parameters and sampling methods
English
English
NoobAI XL: A text-to-image generative model based on Danbooru and e621 datasets
NoobAI XL is an image generation model based on Laxhar/noobai-XL_v1.0. It utilizes the complete Danbooru and e621 data sets for training and supports native tags and natural language descriptions. This model uses v-prediction method (different from eps-prediction) and requires specific parameter configuration (see below for details).
Thanks: I would like to thank my teammate euge for the coding work and many community members for their technical support.
IMPORTANT NOTE: This model works differently than the eps model! Please read the guide carefully!
Model details:
Developer: Laxhar Lab
Model type: Diffusion-based text-to-image generation model
Base model: Laxhar/noobai-XL_v1.0
Sponsor: Lanyun Cloud, Civitai & Seaart
Collaborative testing: LiblibAI, Nieta
How to use:
Method 1: reForge
If reForge is not installed, follow the instructions in the repository to install it; launch the WebUI and use the model as normal.
Method 2: ComfyUI
Please refer to the node example in comfyuiworkflow_sample.
Method 3: WebUI
NOTE: The dev branch may be unstable and may contain bugs.
1. If WebUI is not installed, follow the instructions in the repository to install it.
2. Switch to the dev branch.
3. Get the latest updates.
4. Start the WebUI and use the model as usual.
Method 4: Diffusers
Note: Please ensure that Git is installed and your computer environment is configured correctly.
Recommended settings:
parameter:
* CFG: 4 ~ 5
* Number of steps: 28 ~ 35
* Sampling method: Euler (other samplers may not work properly)
* Resolution: Total area is approximately 1024x1024. Best choices: 768x1344, 832x1216, 896x1152, 1024x1024, 1152x896, 1216x832, 1344x768
Prompt words: (please provide your own prompt words and negative prompt words)
User Guide:
Description: Please provide a clear description of the image.
Quality label:
To ensure that quality labels effectively track user preferences in recent years, we evaluate image popularity through the following process:
1. Standardize data based on various sources and scores.
2. Apply a time-based decay factor based on how recent the date is.
3. Rank the images in the entire dataset based on this process.
Percent Range | Quality Label
------- | --------
95th | masterpiece
85th, <= 95th | best quality
60th, <= 85th | good quality
30th, <= 60th | normal quality
<= 30th | worst quality
Aesthetic tags:
tag | description
------- | --------
very awa | Top 5% of images rated by Waifu-scorer
worst aesthetic | Images ranked in the bottom 5% by Waifu-scorer and aesthetic-shadow-v2
Date label:
Date labels are divided into year labels and period labels. The year label uses the year xxxx format, such as year 2021. Please refer to the table below for period labels:
year range | period label
------- | --------
2005-2010 | old
2011-2014 | early
2014-2017 | mid
2018-2020 | recent
2021-2024 | newest
Dataset:
Danbooru: Latest images as of training date (approximately before October 23, 2024).
e621: e621-2024-webp-4Mpixel dataset on Hugging Face.
Contact information:
QQ group: 875042008, 914818692, 635772191, 870086562
Discord: Laxhar Dream Lab SDXL NOOB
How to train LoRA:
For a tutorial on training LoRA based on sd-scripts, please visit: https://civitai.com/articles/8723
Practical tools:
Laxhar Lab is training a dedicated NoobXL ControlNet model, which is being gradually released. Currently, normal, depth and canny models have been released. Model link: https://civitai.com/models/929685
Model license:
The license for this model is inherited from fair-ai-public-license-1.0-sd at https://huggingface.co/OnomaAIResearch/Illustrious-xl-early-release-v0, with the following terms added. Any use of this model and variations thereof is subject to this license.
I. Usage Restrictions:
Harmful, malicious or illegal activities, including but not limited to harassment, threats and dissemination of false information, are prohibited. The generation of immoral or offensive content is prohibited. Violation of the laws and regulations of the jurisdiction in which the user is located is prohibited.
II. Commercial Prohibition:
We prohibit any form of commercialization, including but not limited to the monetization or commercial use of models, derivative models, or model-generated products.
III. Open source community:
To promote a thriving open source community, users must adhere to the following requirements:
Open source derived models, merged models, LoRA and products based on the above models.
Share work details such as composition formulas, prompt words, and workflow.
Comply with the fair-ai-public-license to ensure derivative works remain open source.
IV. Disclaimer:
The resulting model may produce unexpected or harmful outputs. The user must bear all risks of use and potential consequences.
Check whether the network connection is stable, try using a proxy or mirror source; confirm whether you need to log in to your account or provide an API key. If the path or version is wrong, the download will fail.
Make sure you have installed the correct version of the framework, check the version of the dependent libraries required by the model, and update the relevant libraries or switch the supported framework version if necessary.
Use a local cache model to avoid repeated downloads; or switch to a lighter model and optimize the storage path and reading method.
Enable GPU or TPU acceleration, use batch data processing methods, or choose a lightweight model such as MobileNet to increase speed.
Try quantizing the model or using gradient checkpointing to reduce the memory requirements. You can also use distributed computing to spread the task across multiple devices.
Check whether the input data format is correct, whether the preprocessing method matching the model is in place, and if necessary, fine-tune the model to adapt to specific tasks.
