Hi there and welcome to Celestiblus' sister checkpoint with the idea of having a bit of anime aesthetic, very detailed eyes when upclose and reeling in NoobAI's saturation, if you like this finetune, please consider donating to my <a target="_blank" rel="ugc" href="https://ko-fi.com/bluvoll">Ko-fi</a>, yes this FT is a bit horny, less than Celestiblu, and will make bigger girls, limited compatibility with Loras trained on Illustrious 0.1 and NoobAI 0.5/0.75 but some will work, as this one has custom trained Clip L and Clip GEmphasis mode No Norm is suggested for SDXL models<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/188aa7da-be38-414c-8a96-63dd33708043/width=525/188aa7da-be38-414c-8a96-63dd33708043.jpeg" /><h3 id="-6uq3jngpm"></h3>Huge thanks to <a target="_blank" rel="ugc" href="https://civarchive.com/user/Anzhc">@Anzhc</a> for his <a target="_blank" rel="ugc" href="https://civarchive.com/models/536542/anzhcs-breasts-segmentation-or-yolov8-or-adetailer-model">Anzhc's Breasts Segmentation</a>, <a target="_blank" rel="ugc" href="https://civarchive.com/models/293448?modelVersionId=577989">Anzhc's Face Segmentation (Prototype)</a>, <a target="_blank" rel="ugc" href="https://civarchive.com/models/888145/aaa-anime-sdxl-vae">AAA Anime SDXL VAE </a>(This VAE is already baked on the CKPT) Positive prompt : masterpiece, best quality, Negative prompt : worst quality, low quality, jpeg artifacts, multiple views, 4koma, bad anatomy, bad hands, textPersonal schizo Negative prompt: (worst quality:1.4), (low quality:1.4), jpeg artifacts, scan artifacts, compression artifacts, simple background, bad hands, multiple views, comic, jpeg artifacts, patreon logo, patreon username, web address, signature, watermark, text, logo, artist name, censored, blurry background, Sampler : Euler A AYS, Any Align Your Steps, Euler A CFG++ Align Your Steps, Restart Align Your Steps, Restart KarrasCFG : 4 to 7 (NoobAI feels a bit fried at CFG7+)If you want to support me, consider liking the model, and/or tipping buzz/donating to my <a target="_blank" rel="ugc" href="https://ko-fi.com/bluvoll">Ko-fi</a>, as this was finetuned on 1xA100 Relevant training information:<ol><li>Trained with at least 32k cherrypicked images.</li><li>Total of about ~~900k real steps after batching across 50 epochs.</li><li>Effective batch size was 55 x 1 GPUs x 4 Gradient Accumulation</li><li>Learning rate was 2.7e-6, Clip L was 4e-7, Clip G was trained at 3.5e-7</li></ol>

Files with SHA256: bbf0391ac4745232888a5a96db292580fc4293bf116ebaf693df001091323ec7