🌟 Trained specifically for style mixing on Safe For Work (SFW) dataset.<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/fc81a78b-b565-45cb-80b8-f996963b0410/width=525/fc81a78b-b565-45cb-80b8-f996963b0410.jpeg" />🔹 Text Encoder (strongly reliant on U-Net architecture):<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/067c1ef3-cf14-49ef-b4d2-152e6c69cc03/width=525/067c1ef3-cf14-49ef-b4d2-152e6c69cc03.jpeg" />🔹 Lora Model:<ul><li>Trained using the CivitAI trainer.</li><li>Underwent training for 16 epochs.</li><li>Performance: Lora showcases stable diffusion in XL settings.</li></ul>🔹 Lycoris Models:<ul><li>Configured using <a target="_new" rel="ugc" href="https://github.com/KohakuBlueleaf/LyCORIS/blob/main/example_configs/preset_configs/example.toml">this preset</a> from LyCORIS, but with our custom tweaks.</li><li>Trained on 8xA6000 GPUs.</li><li>Underwent 20-40 epochs with a batch size of 20.</li><li>Module Type Breakdown:<ul><li>LohaModule: 176</li><li>LoConModule: 150</li><li>FullModule: 26</li><li>LokrModule: 700</li></ul></li></ul>🏆 Comparison:<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/f8d5876e-fa9b-4ad9-9d15-c361e48e0c35/width=525/f8d5876e-fa9b-4ad9-9d15-c361e48e0c35.jpeg" /><ul><li>CivitAI's Lora stands out in our tests, but we've yet to explore its full potential in style mixing with other models.</li></ul>Feedback, collaboration, and tests are welcome! 🤟 🥃<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/5abaf673-e7d1-4b5d-a8e2-0c5a042d7405/width=525/5abaf673-e7d1-4b5d-a8e2-0c5a042d7405.jpeg" />

FF Safe AnimXL (Lora, Lycoris +dataset)

Mixed training experiment using | lokr lora lohaMore balanced dims (lower size)module type table: {'LohaModule': 176, 'LoConModule': 150, 'FullModule': 26, 'LokrModule': 700}<pre><code> @unet_target_module = [

 "Transformer2DModel", 

 "ResnetBlock2D", 

 "Downsample2D", 

 "Upsample2D",

]

unet_target_name = [

 "conv_in",

 "conv_out",

 "time_embedding.linear_1",

 "time_embedding.linear_2",

]

text_encoder_target_module = [

 "CLIPAttention",

 "CLIPMLP",

]

[module_algo_map]
 [module_algo_map.CrossAttention] # Attention Layer in UNet
 algo = "lokr"

 [module_algo_map.FeedForward] # MLP Layer in UNet
 algo = "lokr"

 [module_algo_map.ResnetBlock2D] # ResBlock in UNet
 algo = "lora"

 [module_algo_map.CLIPAttention] # Attention Layer in TE
 algo = "loha"

 [module_algo_map.CLIPMLP] # MLP Layer in TE
 algo = "lora"

</code></pre>

Files with SHA256: 337543e6a5d415224161bc3ebb9d47511275be69d2340814e325113815b2cb0b