Brings Valecity's Version of Midna from The Legend of Zelda: Twilight Princess to your renderers! This lora was trained on approximately 150 images with attempts to replicate her classic imp form as well as her more normal Twili form, albeit they have some major limitations at this moment. Regardless I will include the prompts for now and hopefully will make them more reliable in the future. (Any advice is always welcome!)
If you like this, please post what you make below as it helps generate buzz, plus seeing what you make with something I worked on always helps motivate me to keep doing what I do.
As always, please support the original artist for the work that they do. Thank you.
Description
FAQ
Comments (10)
Hm... not my Midna.
No problem at all with that, I just like this version for being a nice hybrid between the two versions, having the tall and curvy look while keeping the same li'l Gremlin vibes
I tried out different things with your LoRa:
Strengths:
+ Images are easier to use in WAN du to Midna being more 'human' than in other LoRas.
+ Without the helmet you can make her very uniquely looking.
+ The style itself is nice and brighter than many others.
Weaknesses:
- Character consistency is really hard, even with the same seed, her patterns always change. This is not good when you want to put videos together from snippets to have a larger video (I would have provided a longer video but that was not possible due to the consistency problem)
- Her face can change pretty fast from 'normal' looking to something rather not Midna looking, especially the nose. I don't know why that happened with certain unrelated prompts mixed in.
Overall it is very solid.
I am not sure if it would be easier to have a better character consistency if the LoRa was divided into a dedicated humanoid and twili form.
Keep up the work! ♥
Yeah, I'll admit this was a bit of a messy one. I had some luck with look changes in the past, and probably overextended it with this one trying three distinct looks. I'm probably going to redo this one, any ideas on how to get her body patterns more consistent? It seems pretty good on the upper torso, but the waist down are a total crapshoot. Maybe by adding a few duplicates from the refsheet pictures to help weight it and add to the learning for that? Thanks for the kind words and like it says above, any ideas and help are welcome!
@StratoKay A bit messy if fine. But it is also really hard for such an character. You would need a huge amount of images with basically perfect character consistency and even then it might still mess up things. A.I. is far more advanced than two year ago but we need at least another leap forward to solve such problems. I would suggest you keep V1 and try out on a more focused model (perhaps even another character who is also equally complex to her). But it is nice that you have shared it with us and I like what you can do with it, albeit better suited for loops.
Don't duplicate images, that messes up your LoRa. Variation is key if the consistency is kept. Check your source material and look if there are variations that may cause it.
On top of that the checkpoint, the base model, can always mess with the output, too. You therefore did good and I see forward to try out other, upcoming LoRas from you.
@FlubbelDon't worry about losing V1, I'm not the type to delete stuff if it's not perfect. I'm also no stranger to over-the-top complexity for character loras, just check out Minerva! https://civitai.com/models/2416563/minerva-spinnerette-v2 Right now, I was working on some style loras at the moment for a change of pace, but I keep an active list of characters that I would like to do so I don't forget. Again, appreciate the kind words, hope you have a good one!
Holy crap this looks so similar to the original version Vale does, you should try doing there OC(‘s) euria (vale-city)
kim (vale-city), if you search them on R34 you’ll see their reference images
Thanks! I strive for accuracy for the stuff I make, admittedly, this one has some issues with the shading derailing with clothes and other things added so I'll probably do a V2. I do like a lot of their OCs, including the older ones, and just added them to my "to-do" list recently after finishing Midna.
@StratoKay Glad to here that there is a possibility of getting a model for those two, the site struggles with making a snake tail that is sentient lol
@TheSloshMaster Little bit janky, but otherwise done!
Edit: Since CivitAI is borked again, here's a link, mate: https://civitai.com/models/2497180/kim-and-euria-valecity-ocs?modelVersionId=2807115
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.