HOME | DD

Description
Heyo, I've been getting a lot of DM's asking for advice, so I posted the main process I use to make images like this:

Basic Process for making ImagesHello. I have been getting a lot of questions lately for advice on how to make images. Here, I will explain my basic process:Pre-requisites:A computer that can run stable diffusion1 model that can do anime1 model that can do photo Many of you don't realize that the starting image I work with usually is very low quality. Here is the example we will be using: ,The amount of LORA that I used had a very detrimental effect on the face. So, what I do is re-sample without any LORA or embedding:,This is already a good improvement, but it's just a 512. The next step is a latent upscale, with controlnet assistance. You're welcome to fix the face before you proceed, but it does not really matter, since the latent upscale will usually un-do any details you produce. At this stage, I swap the model as well: ,Now, we use Ultimate SD Upscale with Tile ControlNet:,Now is the most difficult part: changing the face back to anime, but not making it look bad. I will admit that I still do not have a good way to completely automate this process, and the results will depend entirely on your artistic taste. Here is one that I liked: ,Now, I do some light post processing:,Now, there is just one more thing to fix: the hair does not have a consistent art style. Part of it is photo and part is drawing, so that needs to be fixed with inpainting. I like to do the eyebrows and forehead as well. Basically, leave the face mask blank and just paint around it, ,,That's basically it.Make image with LORAresample without LORALatent upscale w/any controlnet that worksUltimate SD Upscale w/tile controlnetFace fixPost ProcessingSeams Fix on hair/change eyebrows

If you find it useful at all, consider buying the image to show thanks and support!

Description
Heyo, I've been getting a lot of DM's asking for advice, so I posted the main process I use to make images like this:

Basic Process for making ImagesHello. I have been getting a lot of questions lately for advice on how to make images. Here, I will explain my basic process:Pre-requisites:A computer that can run stable diffusion1 model that can do anime1 model that can do photo Many of you don't realize that the starting image I work with usually is very low quality. Here is the example we will be using: ,The amount of LORA that I used had a very detrimental effect on the face. So, what I do is re-sample without any LORA or embedding:,This is already a good improvement, but it's just a 512. The next step is a latent upscale, with controlnet assistance. You're welcome to fix the face before you proceed, but it does not really matter, since the latent upscale will usually un-do any details you produce. At this stage, I swap the model as well: ,Now, we use Ultimate SD Upscale with Tile ControlNet:,Now is the most difficult part: changing the face back to anime, but not making it look bad. I will admit that I still do not have a good way to completely automate this process, and the results will depend entirely on your artistic taste. Here is one that I liked: ,Now, I do some light post processing:,Now, there is just one more thing to fix: the hair does not have a consistent art style. Part of it is photo and part is drawing, so that needs to be fixed with inpainting. I like to do the eyebrows and forehead as well. Basically, leave the face mask blank and just paint around it, ,,That's basically it.Make image with LORAresample without LORALatent upscale w/any controlnet that worksUltimate SD Upscale w/tile controlnetFace fixPost ProcessingSeams Fix on hair/change eyebrows

If you find it useful at all, consider buying the image to show thanks and support!