This Lora replicates the style of the artist Soranamae. <a target="_blank" rel="ugc" href="https://www.pixiv.net/en/users/3000666">https://www.pixiv.net/en/users/3000666</a>It tends to generate NSFW content. Lora responds to the word "soranamae_style". The dataset contained cropped images, so you can add phrases like "head out of frame, white space" to the negative prompt, but it's not mandatoryV2 In this version, there is less noise, and the backgrounds are slightly improved. I recommend generating at a higher resolution (the dataset had a resolution of 1280x1280) and using ADetailer.Flux version To activate the style, simply use the tag soranamae_style Please note that Flux D cannot create NSFW content, so it's recommended to use it in combination with my Lora along with something additional, such as this: <a target="_blank" rel="ugc" href="https://civarchive.com/models/935673/nipplediffusion-flux">https://civarchive.com/models/935673/nipplediffusion-flux</a><a target="_blank" rel="ugc" href="https://civarchive.com/models/935673/nipplediffusion-flux￼Also"> </a>To work with Lora and fine-tuned models, you need to adjust the scaling factor, which is the number written after the ":". This number varies for different models but is likely to be in the range of 1 to 3.5. I added an array in the second post for this model to make it clearer.For example, in &lt;lora:soranamae-style-flux-test2:1&gt;, the scaling is set to 1, meaning it has no amplification. However, in &lt;lora:soranamae-style-flux-test2:3&gt;, the effect of Lora is increased to 300%.<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/a28ddf43-c7b6-43d6-883c-a99f0a94b9bd/width=525/a28ddf43-c7b6-43d6-883c-a99f0a94b9bd.jpeg" />

Style of the Soranamae

soranamae_style-000005.safetensors

1160579_training_data.zip

Files with SHA256: f650ec7a3d775e431e1ae743a0b67ec41b1eac626d86823d427c1792df2f5767