<h3 id="introduction-hlkjl8j2n">Introduction</h3>This is a LyCORIS model for SDXL Anime checkpoints, enhancing the ability of the model to generate pictures of bdsm.<h3 id="usage-g8oki2voq">Usage</h3>Trigger words: bdsm, ropeArtist prompts and style tags are recommended.Recommended checkpoints:<ul><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/332076/kohaku-xl-delta">Kohaku-XL Delta</a></li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/272440?modelVersionId=530170">Heart of Apple XL</a></li><li><a rel="ugc" href="https://civarchive.com/models/260267/animagine-xl-v31">Animagine XL </a>(V2.0 only)</li></ul><h3 id="resolution-5yrlrbnu9">Resolution</h3>This model is trained for resolutions from ARB 1024x1024 with minimum resolution 256 and maximum resolution 1536.<h3 id="training-details-e8i9skujn">Training Details</h3>Hardware: 1x RTX 4090Num Train Images: 1496Num Reg Images: 121Batch Size: 4Grad Accumulation Step: 4Total Epoch: 6Total Steps: 3488Optimizer: AdamW8bitLearning Rate: 8e-5 for UNet (train UNet only)LR Scheduler: cosine_with_restartsCaption Tag Dropout Rate: 0.1Multires Noise Iterations: 6Multires Noise Discount: 0.3Resolution: 1024x1024Min Bucket Resolution: 256Max Bucket Resolution: 1536Mixed Precision: FP16

Anime bdsm Lora

Update the model to V2.1 by adding 1000+ samples for training. Maybe some improvement.

Files with SHA256: 22100e16d854977cfaa41c53694f543939804fa12695e545e0762b51e4d2d259