For Illustrious version of this LoRA: <a target="_blank" rel="ugc" href="https://civarchive.com/models/1683879">https://civarchive.com/models/1683879</a>Should I do Sazer Velsou next just for completion's sake...? I don't really like her character nor her costume design though LOLWould greatly appreciate it if you could read this before posting! <a target="_blank" rel="ugc" href="https://civarchive.com/articles/7060/look-both-ways-before-you-crosspost">https://civarchive.com/articles/7060/look-both-ways-before-you-crosspost</a> Main prompts:<pre><code>sazermithras, white bodysuit, black leotard, belt, armor, white gloves, gauntlets</code></pre>Add for anime images:<pre><code>source_anime</code></pre>Add if doing "full body" or "feet out of frame":<pre><code>boots</code></pre>For helmet with face covered (when doing so, art style tends to change to being completely realistic, probably due to my inclusion of too many photos in the training data set. Helps to add "source_anime" in prompt and "3d, realistic, photorealistic" in negative prompt):<pre><code>helmet, visor, wearing helmet</code></pre>For helmet with face visible (honestly speaking I'm surprised this even works, much less to the degree it has demonstrated to work; I did not provide any of such information in the training data. Could have used the one from the series' transformation scene, but eh, don't want to use the actress's face and ended up having the AI learn to replicate her face. Though sometimes it works wonderfully and at other times it produces a "shaded face" effect due to the visor. Technically it doesn't know what "visor" refers to as there is no data in the training set with "helmet" but no "visor", but you can try adding it to negative prompt along with "shaded face" I guess):<pre><code>helmet, wearing helmet, [eye color prompts], [facial expression prompts]</code></pre>For helmet to be held and not worn:<pre><code>helmet, helmet removed, holding helmet</code></pre>For her weapons, a pair of fans:<pre><code>hand fan, folding fan, holding fan, weapon, holding weapon</code></pre>Armor is removable by taking the following prompts out of positive prompts and adding them into negative prompts:<pre><code>armor, belt, gauntlet, white gloves</code></pre>Additional notes:<ol><li>I had previously assumed that this model would mix well with my previous Sazer Pisces Lora due to me using similar tags for both's trainings, but nope, quite the opposite! When I use both Loras together, it does not know which "armor" and which "black leotard" I am referring to, which results in quite a lot of mixed elements that you would need to RNG brute force through. Not entirely sure if this problem is solvable by using specific tags during training though, so anyone with more understanding of model training can hopefully help to enlighten me</li></ol>

Files with SHA256: eaa45838968b02c9bebb971a072bce7b2ddac35ef7db3295c048f5f36cf9b4c7