This is my first LoRa so if it is unsatisfactory I am sorry I am learning. I hope it turns out great. Thank you for anyone and everyone who will use this LoRa
Gabriella (Angelic Elf OC) is an original character LoRA designed for Anima Base v1.0. She features a hybrid angelic-elven aesthetic with long blonde hair, ice-blue eyes, pointed ears, large cream-colored feathered wings, and a glowing stacked halo.
Trigger Words & Key Tags
* Core Trigger: GabriellaAngel
* Character Traits: long blonde hair, blue eyes, pointed ears, elf ears, white feathered wings, double halo, stacked halo
* Signature Outfit: white toga, asymmetrical dress, gold trim, ruby necklace, gemstone pendant
Recommended Generation Settings
* Base Model: Anima Base v1.0
* LoRA Weight: 0.8 – 1.0
* CFG Scale: 4.0 – 5.0
* Steps: 30 – 35
* Sampler: er_sde
* Scheduler: simple
* Resolution: 1024x1024 (or 2:3 ratio equivalent)
Prompting Tips,(work in progress)
* Maintaining Outfit: Include white toga, gold trim, ruby necklace to lock in her default apparel.
* Changing Outfits: Omit the signature outfit tags and prompt new clothing directly (e.g., gabriella, wearing black leather jacket, blue jeans). Keep white feathered wings and double halo in the prompt to preserve her angelic features.
Description
Comments (4)
Since this is my first LoRa, please give me as much advice as possible. I'll take everyone's advice. I did everything here locally on Civitai. So if there's anything wrong or any advice that you want to give, please give it to me.
I'll give it a few tests to see how it compares to my original. What parameters did you use for training, and what did the dataset look like? I'm very curious to see how you tagged it.
After testing it, it does seem that the dataset you used is rather limited in scope, it really struggles with doing her nude, and has a strong bias for certain poses. Try using a more varied dataset in the future that show off the character in more poses, settings and clothing. A common issue when training small datasets with high step counts.
And I highly suggest you try to be a bit more deriative with future versions of this model. While I am flattered to see someone genuinely like my characters, it's kind of weird and a bit questionable to just willy nilly make your own lora of someone else's OC when they already have made an official one for the character. So, try being a bit different with future releases, use a different name, change the hairstyle, body proportions, hair, eye and skin color, for instance. I'd even suggest bashing together traits from other characters you like to make a character truly your own here, something to set your version apart from the original rather than confusing people looking for the OG Gabriella model.
@ungemozz520thank you for the tips I really appreciate it.