Recommended settings: weight of 0.6-0.9, triggered with yen_w3 for the character, and yen_w3_outfit for her standard Witcher 3 outfit. It works pretty well with DreamshaperXL and base SDXL.
Trained on SDXL with 32 images, 3 repeats, 3000 steps. I'm still trying to figure out the best settings for SDXL LoRA training, so any feedback on that would be appreciated.
Description
Initial Flux version, works best with long descriptive prompts at a weight of ~1.0
FAQ
Comments (10)
can it do a photo if you write "no CGI, no 3dCG" in the prompt?
It doesn't seem to make a huge difference. The lack of negative prompting is a bit of a pain. I might re-train it later with better tagging for CGI/3dCG to see if that makes it more capable of realistic photos.
Sometimes adding "No xyz" can actually makes things worse because it introduces the concept/idea to the AI. Even though you asked for it not to do it. ... Kind of like asking your kids not to do something xD
Impressive. Good job. Thank you.
Feedback:
Why deleting the metadata of the training ? :)
I'm using a tool called SimpleTuner which does not appear to add metadata to the output. Normally I would leave it in, if it exists. I'll upload my training config file in the near future, but I'm still playing around with the settings to find something I like.
@echo_cipher Thx for the help mate !
The community will be happy about it ^^
It's very nice. If you want to do it, please do Liara too. Thank you!
Brilliantly performed by Lora Yennefer, 👏
What was used for the data set of photos from the game and the model taken in the blender?
I ask because I have an ambitin plan to make all the sorceresses out of the universe lora
It would be great if you could update it for flux 2. Thanks :)
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.