Dora trained in re-create Porforever's art style. Is composed of fantasy with cute characters and epic compositions.
I rlly like see your reviews and generations. They are helpful for fix and train better models. Share some hearts at the previews that you like and feel free to leave your comments or requests, but I can't promise anything xd.
Version 2: LoKr
Fixing my mess of training Doras. I can say that comparing them is like comparing apples with helicopters. There is a trigger word, but it completely changes the style of the characters (they don't look chibi at all), so it could be helpful if you only want to keep the color scheme.
Aux tags: porforeverstyl, chibi
Weight: 1
Version 1: Dora
Maybe you'll need to update your Auto1111 to version 1.9.0 or take other steps to make it work in Forge/Confy.
Second try with DoRa's. I increased the initial learning rate now is well present the style in low weights, but maybe now is burned XD, This took more time than the LoKrs, I don't know if it was worth.
Weight: .6 - 1
Description
FAQ
Comments (9)
This works fantastic, no trigger word needed. Doesn't seem burnt to me at all, but maybe it's just due to the model I used? 250 mb is still quite a bit, I'm positive you could train at a lower dim and get the same results, but the style is definitely there.
I'll try with lower DIMs, is just in a previous test I used a lower (with other params) and the results was a little undercooked.
@Konoko Hmm interesting. Maybe I'm wrong then. Perhaps a Dora is that much different from a Lora in that regard
@Light7799 I just confirmed it. I did a trainning for around 9 hours reducing the the DIM to 16, I got a file rlly small, but it doesn't look like the style... needs a weight of 1.6 to looks like... Anyway I'll do a last test before drop it...
@Konoko Huh, how odd. Well no worries then. It gets the effect done after all. I appreciate the effort, but no need to put more stress on your gpu from cooking
@Light7799 Haha this is for the science, just the last test. After that, I'll need remake the other 2 published fail models haha.
@Light7799 Well... I'll drop the Doras XDDD 13 hours wasted trainning another model... I don't know how this one works (Porforever model), I just reduce the dim but I increase the time + lr. Now I have to re-train 3 failed models </3
@Konoko rip, I don't even know if doras are that much better than what we have now with lora, so just not worth the headache and time for sure
@Light7799 they have a good reputation because they works pretty well in the LLMs world, but in SD it's a little different, so it's just wait a little more ultil this gets a better madurity.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.