Style LyCORIS of artist David Balsamique.
Description
Trained on ponydiffusionXL
FAQ
Comments (13)
for pony?
yeah
means that it has DPO, do these loras work for normal XL models? it is not important anyway, since an XL model without DPO is not worth it.
@FireBlack85 They probably only work on any model with ponydiffusion as a base
@MidFR you may be right,in my tests I could not get any normal lora XL to work with the models that have DPO implemented,I am currently using your model a lot,my question is if controlnet XL works with models with DPO,I am looking at how to install controlnet on comfyui to see if it works.
@FireBlack85 pretty sure DPO is a fine tuned 1.5/SDXL model that you merge some block weights in, so most things should work. If you merge with it you're going to lose some of the base understanding from ponydiffusionXL though which is probably not worth it.
PonydiffusionXL is very good by itself I don't think it benefits from DPO. The score prompts kinda function like DPO.
@MidFR Pony Diffusion was the first model to have that technology, Pony diffusion has DPO, it was from there that the other models were born and the ones that are mixed today, that technology came out on January 6 or 7 to be open source and coincidentally the day 7 Pony Diffusion publishes that model when no one else had done it, if you start reading a little on the internet forums you will realize that it is more or less like that, DPO has a good understanding of the language of human preferences and just by writing your ideas you realize the good images that the current models make, expressions, characters, none of that could be done easily without the help of a lora, maybe Pony DIFfusion used training of his images and things that he already had saved but I think the indication improvements are thanks to that. When I asked the creator of Confetti Comrade Mix how he managed to make his model so good, he only told me that he used a model called Comradeship XL. Coincidentally, that model already had DPO and if you You start reading the description of that model, that creator used Pony diffusion to mix his model, your model was mixed with Pony DIFfusion, right? There you have the answer to why your model is so good and coherent, the Comradeship XL v4 was mixed with Animagine XL V3 and it says it has DPO and when you test it, what happens? nothing, it still does the same as the other normal models, but the Comradeship XL v5 was mixed with Pony DIffusion XL X6 and you know what happens? Well, it has very good results, you just have to read the description and you will understand where the magic is.
here for you to read again
https://civitai.com/models/246299?modelVersionId=294219
Here you have the combination that the Confetti Comrade Mix model used
https://civitai.com/images/6319999
the creator of this model himself didn't even know why his model turned out to be so good after mixing it, draw your conclusions
Another model mixed with Pony diffusion was just added a few minutes ago, as you can imagine you already know what results to expect
https://civitai.com/models/301714/cocoamix-semi-real?modelVersionId=338817
Can you upload a 1.5 version? I love this artstyle but unfortunately my GPU can't run XL :(
@MidFR I already knew of that one, but I was curious to see if you could maybe do it better. Nbd tho
Use Forge, it runs smoothly on 8GB of VRAM, and may run fine on 6.
Thank you! This is one of the best lora for pony, gives stable (!) results and when mixed with one of the anime loras can give some interesting outputs. Would be super awesome if it included characters from Behind the Dune!



