NEW: ANIMA (anima base preview 3) version! No trigger word. Supports both natural language prompts, booru prompts and a mix of both (see example images for prompts) Style LoRA trained on on NoobAI-XL vpred 1.0 with images from the artist 2n5 / mikenko2222, most known for drawing traps / otoko no ko's. Version 3.0 does not require any triggerword. older versions do, (jz235) It works on mentioned model and its mixes: <a target="_blank" rel="ugc" href="https://civarchive.com/models/833294/noobai-xl-nai-xl">https://civarchive.com/models/833294/noobai-xl-nai-xl</a> Negatives and quality tags are optional, but are not necessary at all. v3.0 is new and more faithful to 2n5's more modern style. it works well with applying style to things that aren't part of it's training/dataset. I recommend this one. "But mipmup!" You might say. "Why did you train this??? NoobAI already knows 2n5 and there are other 2n5 LoRA's alrea-" NoobAI knows 2n5 already, but its mostly his older works. And the existing LoRA's here on civit suck, with worse training settings and they also probably used most of 2n5's older art when training. What I wanted was a LoRA that could replicate his newer more modern style.

2n5 / mikenko2222 Artist Style LoRA (Anima and Noob Vpred)

<ul><li>No more trigger word needed! Just plug and play. </li><li>Doubled the size of the dataset, including close-up cropped images to capture details better. Total amount of images is 80 (up from 40) </li><li>Pre-downscaled images before training using a new better custom image downscaling script<ul><li>sd-scripts does this already, but I wanted to achieve a higher downscale quality (I can upload it later)</li></ul></li><li>Introduced tag-drop at rates of 15-25% w. protected tags enabled during training, for training the uncond of the model. This is to make the lora's style work better on things it is not trained on (like 1girl stuff). </li><li>Higher batch size/grad acc to generalize better during training.Final lora model is a 50/50 blend of an earlier version with higher amount of tag drop during training and a newer version with lower chance of tag drop during training. the newer version also contains the cropped images.</li></ul>

2n5_v3v7_merge_noobvpred.safetensors

2n5 / mikenko2222 Artist Style LoRA (Noob Vpred)

v3.0 NoobAI_Vpred

v1.0 Anima Preview3

v2.0 NoobAI_Vpred

v1.0 NoobAI_Vpred

Creator: <a target="_blank" rel="noopener noreferrer nofollow" href="https://civitai.com/user/mipmup">mipmup</a>Style LoRA trained on on NoobAI-XL vpred 1.0 with images from the artist 2n5 / mikenko2222, most known for drawing traps / otoko no ko's.<ul><li>Strength: 1</li><li>clipSkip:2</li><li>fgScale:5</li><li>steps:28</li><li>sampler:Euler</li></ul>Version 3.0 does not require any triggerword.It works on mentioned model and its mixes:<a target="_blank" rel="noopener noreferrer nofollow" href="https://civitai.com/models/833294/noobai-xl-nai-xl">https://civitai.com/models/833294/noobai-xl-nai-xl</a>Negatives and quality tags are optional, but are not necessary at all.v3.0 is new and more faithful to 2n5's more modern style. it works well with applying style to things that aren't part of it's training/dataset. I recommend this one."But mipmup!" You might say. "Why did you train this??? NoobAI already knows 2n5 and there are other 2n5 LoRA's alrea-"NoobAI knows 2n5 already, but its mostly his older works. And the existing LoRA's here on civit suck, with worse training settings and they also probably used most of 2n5's older art when training. What I wanted was a LoRA that could replicate his newer more modern style.

Files with SHA256: da4bdc84ad4de94bddd95c02bdd94c5cadc111ef21e4495533bfaafbf59823ca