I'm just experimenting with SDXL.
This time I've reused datasets from my SD1.5 LoRAs.
AetherPunkAI
CoalPunkAI
CogPunkAI
DieselPunkAI
GasLampFantasyAIGasLightFantasyAIManyGaugesAIManyPipesAISteamPunkAI
TeslaPunkAI (+ValvePunkAI)
VictorianPunkAI
LunarPunkAI
BioPunkAI
TumorPunkAI
SovietPunkAI
Results are imho awesome (just check the preview pics)
Perpetual coffee request: Keep me motivated and caffeinated:
Description
FAQ
Comments (8)
The 20 images limit is [...] I need two image posts just to showcase tanks.
Hey! good to see you back,
i also wanted to ask does it take less time to make the loras with SDXL or does it still need a beefy pc?
I tried to make a lora before but my pc couldn't handle it lol
With my current method, it takes more time :/
@konyconi ah, right thank you, I'm really looking forward to seeing more of your SDXL styles!
also do you plan on still making 1.5D loras still? or moving on to SDXL?
yep can confirm what Kony is saying, went from around 30 mins for a Lora to around 2h for me, to get comparable results. Really sad, since SDXL is suprisingly powerful.
@konyconi Currently 3 hours, optimized to 1h and half or 2. Meh, I don't know. Optimized is less quality and it's just 1 hours or 2 more... Umm
@euchale oof man that is still much faster than i expected i think my 1080ti definitely needs an upgrade or im just doing something wrong, takes me near enough 5 hours for a sd 1.5 lora thats using kohya ss.
nevermind just found out im just really stupid... training steps were set to 1000 per image i have no idea how i did that, it dropped to 2 hours now that's about normal now right?
@konyconi Is you method still sensibly the same as before? I'm having trouble finding good kohya-ss parameters for SDXL. Must say I never really tried before SDXL.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.