TRIGGER: h0Sh11
Similar to USA3710. Retagged for better cohesion, retrained for the same, enjoy. (again)
MY SETTINGS FOR ILLUSTRIOUS:
Checkpoint: IllustriousXL smoothitSOLID.safetensor
VAE: fixFP16ErrorsSDXLLowerMemoryUse_v10.safetensors
Sampling Method: Euler a
Sampling Steps: 35
CFG Scale: 5
ADetailer on - FACES v1, Eyeful, Deepfashion
Enjoy what I'm doing?
Liking my models and following me are enough payment for doing this. Reviewing them makes my day. I engage readily with my followers, come say hi.
Description
FAQ
Comments (12)
filename is misnamed on the illustrious version. there's an extra underscore that prevents local A1111 from recognizing it.
even then, i can't really get it to work. how many epochs was this trained on?
I've reuploaded the file itself. Are you having trouble invoking it in your prompt boxes?
As for the usage, my prompt was short, but I'll do some testing with larger ones.
h0sh11, 1girl, solo, ... etc
<lora:Illustrious_Hoshi__Hoshoidk_Style_LoRA:1>
Side note, this is 10 epocs, 653 steps and the same dataset as the pony version. I'm still running mine at 1 str and using smoothftSOLID as my illu model
@insidertrading i renamed the file, still wasn't giving the kind of stuff you posted as an example. i don't really know who hoshi is but this lora's definitely not churning out anything that looks like what he draws nor the examples you posted.
@cobsonlover Mind sharing some settings you're running on discord? insider.trading
@insidertrading that would probably explain it. i'm just using the standard V0.1.
@insidertrading i'm about to post some examples using previous prompts. it's kind of starting to look like his work now (albeit a bit underdetailed) but i'm still not sure how you got the generations that you did.
[... the hell do you mean a moderator has to review my content? get out of my fucking way already]
@cobsonlover I've just submitted a v1.5 of the style with tag shuffling turned on to see if it learns any better. I do remember the dataset for this artist being less than perfect, given they only posted on twitter (fucking compression). We'll see if it does any better.
@insidertrading i don't know what it is. it didn't work on local but suddenly it wants to kind of work on the in-site generator. must be another screwy consequence of using AMD, or maybe it's that illustrious LoRAs don't mix so well with each other at all. i've only been screwing with it for a week
@cobsonlover Me too, I'm on Nvidia running all local, and haven't had trouble getting it to work, but Illu has been a hit or miss on adherence. Willing to help where I can, but I don't understand your local issues without being able to dive into it more. Is your instance not seeing the file or when you go to generate is it not able to invoke the LoRA?
@insidertrading used to not invoke because it didn't recognize filename, then it didn't invoke for some other unknown reasons. try this out with a character LoRA and the colors will seem flat and the proportions a bit wonky.
okay the culprit is DEFINITELY character loras, I don't know what to tell you except to train in as many epochs as possible (i could do like 20 epochs on a character with only 20 reference images without budging the 500 buzz cost, try your luck) to overpower any other lora in such a scenario.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.