It's Tali'Zorah, you bosh'tet!

7/5/2026 Update

Update! I tried a new training process on this lora, hopefully it's better than it's ever been. I've massively cut down the dataset to focus on particularly great art with high resolution and accurate depictions of Tali and her outfit. It's 160 epoch on 63 images on base Anima. I've not used epochs in training much, and that might be like, quite overtrained, which is why I included a weaker version at only 100 epochs. Still, should be a good model. Let me know what you think!

Important Notes:

- The 'standard' version is 160 epochs, I found this to be perfect level for accurately getting her entire outfit, I have included a weaker 100 epoch version just in case anyone wants one that is a bit less overbaked, it's less perfect, but probably plays better with other loras n' stuff. If anyone wants a weaker one, let me know.

- This model is MUCH better at the hexagon pattern on her clothes, use natural language the state that her outfit has it and it should work. I specifically tweaked the dataset to achieve this and it WORKED. 🍾🥂💥

- Anima SUCKS at characters with less than 5 fingers (intentionally anyways lol), so if you want accurate fingers (she has 3 fingers), you might need to do some inpainting for hope for good seed RNG. I didn't not particularly test her weird alien feet, assume the worst... (sorry foot freaks)

- This lora seems to work great with Nova Furry Anima, I would recommend using that if you want somewhat more accurate fingers.

- Yup.

2/17/2025 Update:

Finally got around to updating this... This new noobaiEPSv11-based model has a completely new dataset, going from the previous 119 images to 305 images of Tali! With that totally redone set of images comes with MUCH better tagging, since I ripped them straight from e621 and gelbooru, which is the first time I've done that, so this model supports both e621 AND danbooru tag schemes (about 70% e621). When picking images, I had a major bias to any art that had fine details regarding her outfit, especially the swirly pattern that her purple clothing has, so her outfit is much more accurate now. ☝️

Notes:

• A lot of images have her eyes 'pass through' the helmet. Personally I dislike that, so to prevent it, use "eyeless" in positive and "glowing eyes, looking at viewer" in negatives. (Or embrace it!)

• Noob-based models inherently knows a good bit about Tali, so feel free to play with the weights of this lora. I mostly used 1.0-0.8 weights, but it could probably go lower.

• Tali is the best romance option is Mass Effect.

• Tali has 3 fingers. Using "(3 fingers:1.2), (2 fingers:1.1)" seemed to help with those results. Avoid using just 'fingers'.

• I didn't know this, but loras trained on Epsilon v11 noob work great on the vpred version of the model. Most example images were on vpred. The more you know 🐛

Older Description

3/14/2024 Update:

I updated this model for use with PDXL. Please see the "About this version" section to the right for details. It's pretty much the same model, just trained on better tags, at a better resolution, and with a better optimizer. Hope you enjoy!

Older Description:

This is a lora trained on 119 hand picked images of Tali'Zorah from Mass Effect. It was trained mostly on digital art, but there are some 3d renders in there for variety. I know there is already another Tali'Zorah lora on this site... But I had already trained this before checking so... Here we are! I have not tested between the two, so it's up in the air which one is better. Feel free to test. It was trained on danbooru-style tags, so use those when trying to prompt - I've included the tags in a .txt file in the "Training Images" download below.

It works pretty well with most ckpts, but I recommend Anything 4.5 or AOM3A1B, I used them for the example images.

NOTES:

-It actually seems to understand her alien fingers (She only has 3). I did not do much testing on her feet, but it also seems to kind of get them, too.

-Her weird little skirt(?) thing on her outfit is a bit inconsistent, but otherwise it gets her suit quite well

-A number of images it was trained on had added "eyes" to her, from behind her helmet. If you add "glowing eyes, purple eyes" to the prompt, you can kind of get them, but they may need to be emphasized a good bit.

Example prompt: (See example images, all images should still have their metadata. Just drag the images onto the PNG info tab and import it.)