<h1 id="the-illustrious-version-of-my-chroma-lora">The Illustrious version of my Chroma lora</h1>Trained on 297 images I manually processed (and stole from the internet respectfully. I mean I did it by hand so that's a little better right?). It's extremely flexible and can generate a broad range of sfw and nsfw material. This lora isn't as good as the Chroma one, but that's probably the 6b parameter difference between Chroma and SDXL.<h3 id="and-it-works-with-civitai's-on-site-generator!!!">And it works with CivitAI's on-site generator!!!</h3><h3 id="please-leave-a-like-and-a-comment-if-you-enjoy-my-models!">Please leave a like and a comment if you enjoy my models!</h3><hr /><h1 id="v3.5">v3.5</h1><ul><li>Use it with WAI-illustrious-SDXL: <a target="_blank" rel="ugc" href="https://civarchive.com/models/827184/wai-illustrious-sdxl">https://civarchive.com/models/827184/wai-illustrious-sdxl</a><ul><li>Other Illustrious versions work too. But this is my favourite one.</li></ul></li><li>N works again!</li><li>Sould be better at everything overall<ul><li>Less character bleed</li><li>More variety</li><li>Better at 3D</li></ul></li><li>But the artstyle is less consistent and jumps wildly.</li><li>In comparission to the Chroma version, these optimal settings aren't required to get a usable output!</li></ul><h2 id="v3.5:-comfyui-+-1024p-+-&quot;dpm++-2m-sde&quot;-+-beta_57-scheduler-+-cfg-6.5-+-30-steps">v3.5: ComfyUI + 1024p + "DPM++ 2M SDE" + Beta_57 Scheduler + CFG 6.5 + 30 steps</h2><hr /><h1 id="v2">v2</h1><ul><li>Use it with WAI-illustrious-SDXL: <a target="_blank" rel="ugc" href="https://civarchive.com/models/827184/wai-illustrious-sdxl">https://civarchive.com/models/827184/wai-illustrious-sdxl</a><ul><li>Other Illustrious versions work too. But this is my favourite one.</li></ul></li><li>The Chroma version had N in it too, but Illustrious merged him with Uzi and V. You can still try to generate him with SD-N though.</li><li>In comparission to the Chroma version, these optimal settings aren't required to get a usable output!</li></ul><h2 id="v2:-comfyui-+-1024p-+-&quot;dpm++-2m-sde&quot;-+-beta_57-scheduler-+-cfg-6.5-+-30-steps">v2: ComfyUI + 1024p + "DPM++ 2M SDE" + Beta_57 Scheduler + CFG 6.5 + 30 steps</h2><hr /><h2 id="prompting:">Prompting:</h2><ul><li>Works best if you start the prompt with the character tag: SD-V, SD-J, Uzidoorman, Cyn</li><li>If a character isn't generated correctly, type more describing traits: SD-V -&gt; SD-V with short white hair, white skin<ul><li>There can be mixups with a too short prompt.</li><li>Illustrious loves to give the drones human skin. Fix that by typing: white skin</li></ul></li><li>The Lora likes to forget the tail. Luckily you can remind it by typing: tail</li><li>Natural language prompts work best, but can be combined with tag based prompting</li><li>Sometimes the negative prompt can mess things up. Keep it minimal!</li></ul><hr /><h2 id="bias:">Bias:</h2><ul><li>The Lora is heavily biased towards generating V, J and Uzi. N and Cyn need a few more describing words.</li><li>Most APIs I used to get images are on the more nsfw side of things so... write "wearing clothes" to hide that bias.</li></ul>

[Illustrious] V, J, Uzi, Cyn (Murder Drones)

<ul><li>Better Character consistency</li><li>Better composition</li><li>No "white skin" anymore! (most of the time)</li><li>Better at 3d style</li><li>N is back!</li><li>Less overlapping with the triggerwords</li></ul>I tried to enhance performance with multiresolution training and lower batch size and it went pretty well.

Files with SHA256: fe13ddaeef9e732a1bc0b3c1e59d4e6cff46a29d1ce9e69892944314804f7a8b