<h2 id="lora:-accelerando's-girls">Lora: Accelerando's Girls</h2>Tamaki has always been one of the most iconic hentai character of the early 2010s, so i thought that it was unacceptable that anima would recognize her and her peers. This is what brought me to train the lora you shall see in this collection.<h3 id="how-to-use-it">How to use it</h3>Generate an image normally with your prompt and seed (<a target="_blank" rel="ugc" href="https://huggingface.co/circlestone-labs/Anima">FOLLOWING THE ANIMA GUIDELINES</a>) and adding the activation tag for the relative lora character (you can find it in the lora version info).Then if you want a style closer to the original i suggest you to remove "year 2025" and such tags from your anima prompt and add "year 2010" and "year 2009" in the positive prompt.PS: Specific Version notes might be added in the version note.<hr /><h1 id="versioning-log:">Versioning LOG:</h1><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/226f1f5a-454e-4368-82c0-61fa68044192/original=true/226f1f5a-454e-4368-82c0-61fa68044192.jpeg" /><h3 id="hirohashi-(v.1)-06072026:">Hirohashi (v.1) [06/07/2026]:</h3>This is the lora of the serie featuring hirohashi, right now you can expect:<ul><li>A good understanding of the character.</li><li>Her classic serafuku, can be called via "serafuku" tag</li><li>A style close enough to the one of the serie.</li><li>Enough flexibility to experiment with average scenarios and with some tweaking even in complex ones.</li></ul>This is the forth and last lora of the serie. We wanted to have one fore rokoko and kanue but both of them have not enough screen time to construct a decent dataset to train on. (we could use synthetic data, but not in the near future).<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/379471dc-a545-442a-a57d-f3b71b95aa78/original=true/379471dc-a545-442a-a57d-f3b71b95aa78.jpeg" /><h3 id="tomoe-(v.1)-05072026">Tomoe (v.1) [05/07/2026]</h3>This is the third lora of the serie featuring tomoe, right now you can expect:<ul><li>A perfect understanding of the character in both her version, mature and young.</li><li>A style similar to the one of the serie can be obtained with this version.</li><li>Enough flexibility to generate complex images, this might require some tweak of the lora strenght for really complex ones.</li></ul>We can consider this version the best one of this lora set yet. It include both ideas of the character, both her canon outfit can be obtained via keywords, without any unnecessary concept bleeding.As of now, we are discussing, better approaches to obtain more lora like this and less like the ayumi one. We might also decide to refactor our model descriptions and the way we publish LoRA to convey better info, and set up a schedule for releases.<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/99db08ca-8513-41b4-8e3f-b2fdff190f97/original=true/99db08ca-8513-41b4-8e3f-b2fdff190f97.jpeg" /><h3 id="ayumi-(v.1)-05072026:">Ayumi (v.1) [05/07/2026]:</h3>This is the second lora of the serie featuring ayumi, right now you can expect:<ul><li>A rough understanding of the character and her main outfit</li><li>Enough flexibility to use them for simple images</li><li>A model that even with an undersized dataset has a medium quantity of bias</li></ul>We consider this version a failure, but we share for completeness sake, a retrain is expected in the distant future after acquiring enough material to improve the original data.<h3 id="tamaki-(v.1)-04072026:">Tamaki (v.1) [04/07/2026]:</h3><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/0e625278-561f-43d5-913e-7ab4f296cdc2/original=true/0e625278-561f-43d5-913e-7ab4f296cdc2.jpeg" />This is the first lora of the serie, featuring tamaki. Right now, expect:<ul><li>A solid understanding of the character</li><li>Enough flexibility to use both her canon outfit and other outfits.</li><li>A model that has a low quantity of bias</li></ul>Other models are planned (ayumi shall be the next).<hr /><h2 id="disclaimers:">⚠️⚠️⚠️ Disclaimers: ⚠️⚠️⚠️</h2><ul><li>Built on the ANIMA architecture. ANIMA is still evolving, so full compatibility with future versions isn't guaranteed, if something looks off, it's probably version drift, and you might need to swap to a newer version or wait for an update of the model.</li><li>Our train capabilities are limited and so is our testing environment... we acknowledge that behavior may vary across models and settings. Use it consciously.</li><li>We do not cherry pick our outputs, we do not use upscalers or adetailer for our gallery images and we won't ever do it, because we want to show you the quality for the average output you might get.</li></ul>( PS. we might use the techniques and the tools mentioned above on later posts to show the full model capabilities, and for our entertainment)<hr />Thanks for using those models. We'd like to see your creations underneath the model to help us understand what we did wrong and what we can do better. This is all we need to make the next model even better.

Accelerando's Girl Collection

This version include the character Tamaki from the accelerando serie. The lora itself was trained on a dataset of 66 images.<pre><code>"tamaki (accelerando)" ------------------- for booru tag prompting (stable)
"tamaki from accelerando" ---------------- for natural language (stable)

"antenna hair" --------------------------- for having the 3 ahoge (unstable)

"white shirt, bowtie,
green apron, black skirt" ---------------- for the main outfit (stable)

(year 2009, year 2010) ------------------- for a retro style (stable)
 </code></pre>

Files with SHA256: 4e86e9ea78b885631e6b13def1603a7e6a091f90eb9e24c52f5ac27c85d2f4cb