My Illustrious-based Nanoha model is now available at NanoMix ~Welcome to the Nanohaverse~!
This LoRA has been trained on Official Art of the Magical Girl Lyrical Nanoha franchise, focusing on the protagonist Takamachi Nanoha, and has been trained on Pony V6 XL and will not work on other models, or in conjunction with other LoRAs. The intended purpose of this LoRA is to generate images of Nanoha in the Official Art style, meaning that this LoRA affects both character and style.
The activation tag is takamachi nanoha and the recommended LoRA strength is anywhere between 0.7 and 1.0, with 0.85 serving as a good starting point. Recommended additional tags include small breasts, twintails and white ribbon or black ribbon. An interesting tag to experiment with is anime screencap. All testing was performed using the common prompt score_9, score_8_up, score_7_up, source_anime.
Description
FAQ
Comments (12)
Cute! Gonna give this a download. 😎👍
Is Pony the new hotness or something? I've been kind of out of the loop with models, but I see a lot of Pony Loras these days. Can't deny the results though. Those previews look great!
Thank you! Good to see you're still around, fellow Nanoha-fan! Pony is definitely surging in popularity right now. In my opinion, it's the first SDXL model that reaches a quality that makes it worthwhile to learn and use over the well-established SD1.5 models. The adherence to character appearance and overall style, combined with the deep understanding of intentions and prompts, is nuts. But I hope to see more anime-tailored models to follow soon, like Hassaku XL, and train the best waifu all over again.
It's become something of a habit to scroll through new models in the morning, yeah. I make sure to collect any good ones, Nanoha and other, for future use! She'll probably be right at home in that Helldiver outfit I saved the other day, haha.
I see. So Pony is a growing XL variant that could serve as 1.5's successor. I was wondering when it would happen. Got distracted by the announcement of SD3, but it looks like I'll have something to play with first. Does it still need ControlNet, or can I be a bit more liberal with it?
Thanks again for continuing the valiant work of training these models! They're such gems.
@Ichibey You can be far more liberal with it. SDXL models have far better understanding of anatomy and a much better understanding of natural language. Almost everything I throw at it comes out great. I rarely need ControlNet now, except when dealing with multiple characters (and Nanoha doesn't play well with other characters yet).
@Deathspike Been getting into Pony as of late and you're right. I actually tried ControlNet with it, and it actually degraded the image quality while at the same time not being as strong in influencing the generation, even when I had it on max. I must be doing something wrong idunno. But either way I'm not feeling like using CN these days outside of 1.5
Thanks for enlightening me on this development! Now I can wait a few more months for SD3. :)
@Ichibey When SD3 releases, we'll get to figure out how to train a whole new Nanoha yet again!
@Deathspike So true. I got on Pony's discord, and it seems v7 is being prepped for SD3. Looks like this is really as good as it'll get until then. But I bet that next gen Nanoha will be quite a showing!
its amazing how much you can reproduce without Negative Prompt.Great job!
I'm so happy to see that you finally made a nanoha lora for pony, I tried to make pics of girls in nanoha series with prompts only using ponyxl, but the results are not good. Now I'm looking forward to more nanoha loras for pony. So many thanks.
17 pictures in dataset is not many. The head style is often a little distorted. Hope for next version.
Quality over quantity. There will likely not be a second version, but feel free to continue training with the open-source dataset. Just mind the license.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.