Komori Kinoko ~ Invisible Girl
Thanks for checking out my LoRA. Give it a try and if you like it, leave a 👍
Again, another character with not a load of official artwork (albeit more than invisible girl for obvious reasons). This character is super cute. The main challenge here is that most of the time she appears in the same costume, so you'll likely spend a lot of time fighting the LoRA for control. I may go back and add more reference from the anime for better outfit control, but alas I must continue on with other projects.
This character is a bit more strict than Toru is. You may need to re-roll a couple of times. I've had probably a 40% - 60% success rate with my prompts.
Usage
Suggested Model: PonyXL v6
Suggested Weight: 0.6-1.0
Trigger: komori_kinoko
Optional tags: red dress, hat, polkadots
It should work decently without the optional tags, but I find it more consistent with them.
Outfits
Komori really only had one costume in her training set. Luckily, you can prompt for some alternative costumes. See some of the examples for ideas.
Tips
Sometimes her hair will get longer if you have her cosplaying as other characters. You can reduce this with "short hair" or "medium hair" tag.
The character will default to large breasts. If this is not your preference, you'll need to account for it in the tags.
You must specify that the dress is red to get her actual costume. For some reason, sometimes the dress will default to like a navy blue.
Description
First version!
FAQ
Comments (6)
great work again! though if you ever do a V2 may I suggest changing the tag for her mushroom hat from her hero outfit from hat to specifically mushroom hat? I noticed it kept overwriting when I tried to give her a sun hat for a couple outfits I was putting her in
Great idea! Yeah, her hat was suuuuper finicky. I believe in the dataset there are some images tagged with mushroom hat specifically, while others just use hat. I think reducing the redundancy on a revision would help a lot. Thanks for checking out another one!
@bzbub no problem 👍
Very nice model but the file size is a lil bit too big
Noted! I have a beefier computer so I train models with a slightly higher rank for accuracy. With future models I may add a "lite" version trained on lower rank to reduce file size at the cost of accuracy. I'm still new to training LoRAs so I really appreciate the feedback! :)
I see you have a few models under your belt so please let me know if you have any tips!
@bzbub You want your Network Dim to be either 32 or 16 depending of how many pics you have on your dataset, the less the lower, while your Network Alpha should be half or your Dim (16 or 8)
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.