Get early access to build and test build, be able to try all epochs and test them by yourself on Patreonor contact me for support on Discord
To support me you can buy me a coffe at https://ko-fi.com/antonioriolo
Training settings:
35 images, 20 steps, 15 epochs, you find many epochs (with test images using plot xy script) on my Patreon while supporting me.
Recommended settings:
Make sure you are aware on the usage instructions of LORA
Sampler: DPM++ SDE Karras (Recommended for best quality, you may try other samplers)
CFG Scale : 5 to 10
LORA weight for txt2img: reccomended 0.4-0.7.
Keys: human on fire, horse on fire, animal on fire, angel on fire, skull on fire, red fire, blue fire
This is a very hard to use model. In my experience, to generate the images i posted i had to try a lot of stuff, changing weights and doing many plot xy.
I'm going to post epoch 10 but as far as i could see different prompt need different epoch to work well with, in fact many of the images are with different epoch so you will not be able to replicate it excatly by copying the prompt and using another epoch.
Wrong weights or wrong epoch will give artifacts or no fire at all.
Doesn't work well with many loras in it, the only lora i managed to let it work with is my Luxury Dress model that you will find in many images posted.
Most of the training images have monochrome background, it's not easy to make it generate an image with the background you want, not without artifacts.
Considering all of this, if you manage to make it work it can generate amazing images, i'll wait a good prompter to post great images about this model, i'm very curious.
And remember you can find all epochs in my Patreon, epoch 10 gave best result in most images but sometime a more trained or a less trained epoch gave best results, if you like the concept you should try them all!
Enjoy!
If you'd like to support my efforts and have access to early and bleeding edge builds , please think about joining my Patreon: https://www.patreon.com/betterCuteAsians
评论