Update: By request and popular demand, now available in a PonyXL Variant.
Adaptation of my SD15 grimacing LoRA to XL. First attempt at migrating the facial expression series, would love feedback on how well it works comparatively.
Pony Info:
Works really, really well with Hyper, but might need a little more tuning down on "realism" models. In particular you should be careful with embeddings or LoRA's which enhance the "photo" nature of the generations, since together they oversaturate pretty easily. It can be helpful to use adjacent words like clenched, or wincing, rather than the trigger word.
Normal animated pony models can mostly already make this expression, but this LoRA can help preserve it in scenes where there's a lot else going on. Using adjacent words (rather than the trigger) can help mitigate influence on style.
Starter Realistic Pony Prompt
photo of <subject>, clenching
Vanilla SDXL Info:
Starter Prompt
<subject>, with a pained grimace
Warning that I never cleaned up this dataset, so the images are a little horny by default unless you prompt them otherwise.
Description
Same base training structure as the XL model, with Pony Base.
FAQ
Comments (3)
Are you planning on updating all your facial expression loras?
Maybe, but probably not unless I can come up with a more consistent process.
The source data for most of my LoRAs was barely high enough resolution to support 512x512 training. Training this one required a pretty high-effort process of carefully upscaling all of the images slightly differently to prevent wiping out the nuances and fine details.
There's probably a more creative way to train XL / Pony that's more native to those models and would work without mostly replicating what I was doing in SD15, but it's a massive amount of extra iteration time to train and generate test images (both take 10x longer).
I'm kind of lukewarm on XL and Pony anyway. XL seems to encode so much detail that it gets wonky really, really fast, and the sweetspot for training isn't clear. I still haven't been able to produce a Subject/person LoRA that's anywhere near as good/consistent as SD15. Pony is really, really impressive for 2D stuff, but I'm extraordinarily skeptical any of the realistic models are ever going to get past the plastic doll phase. I'm hoping that there'll be another Pony-magnitude breakthrough, but until then I'm more interested in experimenting rather than trying for breadth.
@thegipper do you think you could share your SD 1.5 datasets for all the other emotions? I know you said they were 512x512 but honestly I'd love to (and I'm sure others would) take a crack at it and try to create more SDXL versions of them. I've never done Theme/concept loras before but I'd love to see how well I could do.
You make the best emotional loras in all of civitai so that's why I want to see if I can piggyback off your high-quality work.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.