Polar Fox
Well-Known Member
- Feb 20, 2022
- 1,824
- 9,395
These are the images I created last year (mostly). Now I use Flux. Upscaling old images with Flux using img2img method works great. I have never created a negative embedding because I don't know how to do it. Kohya has been giving me an error for a while. Which embedding did you develop among the ones I use? By the way, here is my civitai link:Thank you, i will try. But really denoise 0.2? is upscaled?
Also glad to see you use one of my embedding in negative.
Really qwesome! Freckless are good, only eye is the problem. is because upscale?
And like you, i made myself model, lora and embedding. Had same problem with eye in some merge, but correction is possible (i had it good now).
Pretty much - although Civitai's online generator does allow the use of LoRAs, it's only those that are already on Civitai. Also, there are a lot of advanced features in Forge/A1111/Comfy that aren't present in any of the online generators (presumably to stop you breaking them!)Looked through a couple of the pages on here, and some of the results are very, very good. But from looking (briefly) at places like Civitai and that, I'm not seeing any online service or tool that will both let you create NSFW, and has good character persistence. IOW, if I want the same character in the living room, bedroom, and bathroom - with a couple poses in each - it seems to be somewhere between difficult and impossible.
I'm guessing that from what I have been able to pickup that some of you are getting decent results by downloading a model, training your own Lora, and then doing local generations. Is that correct? Is that also pretty much the only way to get consistency right now?
Thanks for the answer and the direction!Further discussion on this sort of thing (and lots more) is over on the SD thread