<h1 id="kirazuri-lazuli-(noobai-v-pred)-i86hl9fyv">Kirazuri Lazuli (Noobai V-Pred)</h1>This checkpoint is a personal project trained locally from <a target="_blank" rel="ugc" href="https://civarchive.com/models/833294/noobai-xl-nai-xl?modelVersionId=1190596">NoobAI-XL (NAI-XL) V-Pred 1.0-Version </a>on a 4090 over one month on a small dataset of 12,561 total images.It focuses on adding additional knowledge since the data cutoff of the base model (2024/10/24), including styles, concepts, and characters from anime, video games, and virtual youtubers.<h2 id="usage-u795q3iwf">Usage</h2>Previews are generated with a ComfyUI workflow using DynamicThresholding, Upscaling, and FaceDetailer.Forge/Re-forge can also be used.For samplers, recommendation is Euler for generation, Euler Ancestral for upscaling/inpainting.<h3 id="recommended-prompt-structure:-4fw6z8mbb">Recommended prompt structure:</h3>Quality modifiers <code>masterpiece, best quality, very aesthetic</code> should be positioned at the end of the prompt.Artist names can be prefixed with <code>artist:</code> to prevent token bleeding with artist names and concepts.A1111 schedule prompting syntax is used in ComfyUI through the <a target="_blank" rel="ugc" href="https://github.com/asagi4/comfyui-prompt-control">comfyui-prompt-control</a> extension to combine artist styles, i.e: <code>artist:[artist1|artist2|artist3]</code>Positive prompt:<pre><code>{{characters}}, {{copywrites}}, {{artists}},
{{tags}},
absurdres, masterpiece, best quality, very aesthetic</code></pre><h2 id="training-details-s6vgvd4lj">Training details</h2>Full kohya-ss training configs can be found on <a target="_blank" rel="ugc" href="https://github.com/motimalu/diffusion-training-configs/tree/main/sd-scripts/SDXL/adafactor-01-full-fp32-v-pred">github.</a>Iterative checkpoint training approach inspired by <a target="_blank" rel="ugc" href="https://civarchive.com/models/141592">PixelWave.</a>This involved training in dataset batches of ~1200 images, for 10 training sessions, before finishing with an 11th aesthetic finetune dataset of 267 images.<ul><li>Adafactor optimizer</li><li>Full-fat fp32 training precision</li><li>Batch size and LR were adjusted multiple times<ul><li>Batch size 4, LR 6e-6 seemed most stable</li></ul></li><li>TE trained for the 10th and 11th training sessions at Batch size 4, LR 2e-6</li><li>Regularization dataset generated from the 10th checkpoint used in the final aesthetic training to preserve the previously learned characters</li></ul><h2 id="recognitions-s5uhld8rk">Recognitions</h2>Thanks to <a target="_blank" rel="ugc" href="https://huggingface.co/Laxhar">Laxhar Lab</a> for the <a target="_blank" rel="ugc" href="https://civarchive.com/models/833294/noobai-xl-nai-xl?modelVersionId=1190596">NoobAI-XL (NAI-XL) V-Pred 1.0-Version</a> base model.Thanks to <a target="_blank" rel="ugc" href="https://github.com/narugo1992">narugo1992</a> and the <a target="_blank" rel="ugc" href="https://huggingface.co/deepghs">deepghs</a> team for open-sourcing various training sets, image processing tools, and models.<h2 id="license-foxpa8hxt">License</h2>This model's license inherits from <a target="_blank" rel="ugc" href="https://huggingface.co/OnomaAIResearch/Illustrious-xl-early-release-v0">https://huggingface.co/OnomaAIResearch/Illustrious-xl-early-release-v0</a> <a target="_blank" rel="ugc" href="https://freedevproject.org/faipl-1.0-sd/">fair-ai-public-license-1.0-sd</a> and adds the following terms. Any use of this model and its variants is bound by this license.<h2 id="i.-usage-restrictions-l4bae0qyp">I. Usage Restrictions</h2><ul><li>Prohibited use for harmful, malicious, or illegal activities, including but not limited to harassment, threats, and spreading misinformation.</li><li>Prohibited generation of unethical or offensive content.</li><li>Prohibited violation of laws and regulations in the user's jurisdiction.</li></ul><h2 id="ii.-commercial-prohibition-8czhb7ilf">II. Commercial Prohibition</h2>We prohibit any form of commercialization, including but not limited to monetization or commercial use of the model, derivative models, or model-generated products.<h2 id="iii.-open-source-community-hl88kkeyl">III. Open Source Community</h2>To foster a thriving open-source community, users MUST comply with the following requirements:<ul><li>Open source derivative models, merged models, LoRAs, and products based on the above models.</li><li>Share work details such as synthesis formulas, prompts, and workflows.</li><li>Follow the fair-ai-public-license to ensure derivative works remain open source.</li></ul><h2 id="iv.-disclaimer-8k57px2rb">IV. Disclaimer</h2>Generated models may produce unexpected or harmful outputs. Users must assume all risks and potential consequences of usage.

Kirazuri Lazuli (Noobai V-Pred)

Trained on <a target="_blank" rel="ugc" href="https://civitai.com/models/833294/noobai-xl-nai-xl?modelVersionId=1190596">NoobAI-XL (NAI-XL) V-Pred 1.0-Version</a>Dataset cutoff: 2025/05/25

Files with SHA256: 655239b43a8cd98bf1ad09ec0ba9a8d531171549e59cc25a78ecb8973fb6dcfe