Hello everyone,Today, I’m posting the last LoRA from the anime X/1999 (also known as “X”). The final character is Toru Magami.<h3 id="version-1.03-only-uses-original-anime-screenshots-instead-of-images-generated-by-another-model-so-the-character-looks-closer-to-the-actual-design.-the-downside-is-that-the-model-may-lack-detail-or-flexibility-due-to-the-small-amount-of-training-data.">Version 1.03 only uses original anime screenshots instead of images generated by another model, so the character looks closer to the actual design. The downside is that the model may lack detail or flexibility due to the small amount of training data.</h3><s>To be honest, one particular thing made her really hard to complete: the multiple scrunchies on her ponytail. It turns out that defining a specific number of the same object in a single image is quite difficult. The model would usually generate either just one scrunchie or a random number of them.</s><s>To show you how tricky this was:</s><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/e7126b66-59cf-4d6c-addf-48ab5ded19a3/width=525/e7126b66-59cf-4d6c-addf-48ab5ded19a3.jpeg" /> 🤣🤣🤣<hr />Prompt: look at "Trigger Words" on the right side 👉Changelog: look at "About this Version" on the right side 👉<h1></h1>My Settings <a target="_blank" rel="ugc" href="https://github.com/Haoming02/sd-webui-forge-classic">(Forge Classic</a>)*:*note, the example images only use high-res fix.Checkpoint: waiNSFWIllustrious_v160Sampler: Euler aScheduler: AutomaticSamplings Steps: 30Hires.fix: R-ERSGAN 4x+Anime6B, 20 Hires steps, Denoising Strength: 0.5Extension:Adetailer (highly recommended)

Toru Magami - X 1999

This version only uses original anime screenshots instead of images generated by another model, so the character looks closer to the actual design. I also used a new dataset based on the 4:3 version of the anime, so the characters are no longer squished.

Files with SHA256: 230de432e7331e515edf8c598651d79aa8645a03671d6d960bedbf63198863b7