Please read through the entire description (might need to be expanded) and the version change notes as they cover a lot of information about basic use cases and limitations. The description is always focused on the latest version. Thank you!This is an SD 1.5 LoRA for the character Cynthia / Shirona from Pokemon. It works for models based on SD 1.x only, it will not work for models based on SD 2.x or SDXL!Example images were picked from pure 512x512 txt2img results and then re-created at 1024x1024 using txt2img with moderate Hi-res Fix settings (upscaler Lanczos, denoising strength 0.4). This improves faces and other details that are almost impossible to get correctly and consistently at 512x512 (limitation of the technology) while still giving a realistic impression of what it looks like. Pure 512x512 results will have more distorted faces and less detail.Please see the version change notes for the training and example image generation models as well as the used weights as they might change between versions. Remember that you might need to adjust weights to best suit your use case!Remember to add the tag cynthia \(pokemon\) (with backslashes intact) to your positive prompt.The training set contained multiple variations of Cynthia's signature look so you might need to put combinations of the following tags in your positive or negative prompts to get the desired results. Some should probably be used together (separated by plus in this list):<ul><li>fur collar</li><li>fur-trimmed coat</li><li>hair ornament</li><li>hair over one eye + one eye covered</li><li>high heels</li><li>pants</li><li>shirt + v-neck</li></ul>As the training set contained a few images with floating hair, you might need to add floating hair to the negative prompts if not desired.Depending on what you want to generate, you may also want to experiment with adding cleavage to the negative prompt or specifying a breast size as they vary quite a bit.Known Limitations / Problems:<ul><li>The shoes were not present in a lot of images and not a focus for the training so they sadly will not be consistent at all.</li><li>The gems on the coat also sadly seem to be quite inconsistent despite being in a lot of images.</li><li>This model seems to be much more fixated on the simple backgrounds used during training than others I've done (despite some variation in the training set). You may want to increase the weight of background tags and/or add simple background or similar to your negative prompts.</li><li>It is also sometimes a bit fixated on the 2-3 images in the training set where a covered navel was visible under the shirt so try covered navel in your negative prompts.</li><li>Despite a lot of attempts, the separation between what is part of the coat vs. what is part of the shirt below also does not seem to have stuck properly yet.</li><li>The coat + shirt seem to get combined to a (suit) jacket fairly often so try jacket or suit in your negative prompts if that happens.</li><li>From behind, the pants tend to come out as jeans, so try jeans in your negative prompts there.</li><li>From behind if not prompting for the coat, the shirt sleeves tend to have markings where the fur trims would usually be. I have no idea how to fix this yet. Maybe try lapels in the negative prompt?</li><li>The hair can be wildly inconsistent, especially from behind (but that's also the case for every illustration of Cynthia I know of).</li><li>The hair color might sometimes bleed into the sleeve color. I'd just regenerate in that case.</li></ul>

[SD 1.5] Pokemon - Cynthia / Shirona

General changes:<ul><li>Removed unnecessary general tags coat and shoes (they already had more specific equivalents, see description)</li><li>Changed example weight from ~0.5 to ~0.7 (result of training changes)</li></ul>Changes to the training process (with the goal of hopefully making this and my other models more flexible and compatible while reducing overfitting):<ul><li>Switched from a flat 200 repeats x 1 epoch to 20 repeats x 10 epochs (max) and selecting the best results</li><li>Lowered network training alpha from 128 to 64 to hopefully increase the quality of the result a bit</li><li>Resized LoRA from rank 128 to rank 64 rank after training (still training with rank 128) to reduce size and "smooth out" some of the training results while not changing much (going down to rank 32 seems to have too much influence on the quality of results, so I kept it at 64)</li></ul>Used weight: 0.7Training model: <a target="_blank" rel="ugc" href="https://civitai.com/models/66/anything-v3">Anything V3</a>Example image generation model: <a target="_blank" rel="ugc" href="https://civitai.com/models/4451/abyssorangemix2-hardcore">AbyssOrangeMix2 - Hardcore</a>

Files with SHA256: 53c257917523149893bc4336170289956553dc4c89b1674a31c8a2828d7ea222