Unfortunately all the skin details are just gone and we are back straight to the 1999 when all photographers just used to smooth the skin with the Gaussian blur filter 😂
Really good question, there’s a couple ‘redhead’ Lora files on civitai which might do the trick. Using the word freckles in your prompt. Inpaint the freckles onto your image. You will need to upscale it and do a img2img to increase the detail on it if they come out blurred. I haven’t tried it out myself yet so I can’t speak to how realistic it will come out. Perhaps start with a low number like 0.3 or 0.5 on the Lora to see what it does. Sometimes they go too far when set to :1
Thanks for this video man. Lately I've been using stable diffusion to enhance my photography and man I kid you not this is a huge game changer. I hope many more photographers start doing things like this
@@sebastiantorresvfx Indeed. I cant afford photoshop atm so this is a great alternative. Seriously thank you for this. I used your method with the Photonv1 model and man oh man this is such a game changer
To be honest I'd say that is too much retouching. The face is passable but the body is far too much and looks incredibly fake. Granted your final step was just a quick example but still. Furthermore how are you going to retouch anything larger than 1024 for say print purposes since you can't upscale the photograph in SD. Perhaps I'm missing something here but there are a ton of better Photoshop specific tools for exactly this purpose. Not bashing SD, rather saying use the right tool for the job.
Hi Marchant, to your first point, yes I kind of rushed through the body for this example but you are able to take your time and generate multiple versions then mix and match the best generations to get more detail or less detail depending on what you’re after. As for the 1024, when you’ve upscaled the image and set the inpainting to masked only it fits the 1024 pixels into what ever your masking. So keep the masks at roughly that size. If your upscaled image is 4096/4k don’t make the masks more that a 4th the size of the image. I hope that makes sense. As in do small chunks at a time. That way you retain the image quality. I agree there’s other ways to do this using photoshop. This is just another tool in your arsenal to get the job done.
Very informative. Thank You. Is there a chance you could share your Stable Diffusion upscaling workflow?
I’ll be doing a live stream soon, I can touch base on that during the stream.
Unfortunately all the skin details are just gone and we are back straight to the 1999 when all photographers just used to smooth the skin with the Gaussian blur filter 😂
amazing! congratulations!!! very good!
Thank you 😃
Great video. How do you do the opposite? I love freckles but I can’t get stable diffusion to generate realistic looking freckles.
Really good question, there’s a couple ‘redhead’ Lora files on civitai which might do the trick. Using the word freckles in your prompt. Inpaint the freckles onto your image. You will need to upscale it and do a img2img to increase the detail on it if they come out blurred. I haven’t tried it out myself yet so I can’t speak to how realistic it will come out. Perhaps start with a low number like 0.3 or 0.5 on the Lora to see what it does. Sometimes they go too far when set to :1
@@sebastiantorresvfx I’ll give that a try, thanks
Thanks for this video man. Lately I've been using stable diffusion to enhance my photography and man I kid you not this is a huge game changer. I hope many more photographers start doing things like this
I agree it’s only a matter of time, now with photoshop integrated Ai generator it’s another tool on our belt.
@@sebastiantorresvfx Indeed. I cant afford photoshop atm so this is a great alternative. Seriously thank you for this. I used your method with the Photonv1 model and man oh man this is such a game changer
Don’t forget you can use completely different models when inpainting also. Very handy technique.
@@sebastiantorresvfx Of course!
i am, yours subscriber numbers: 100 . congratulations 💯
Wooo! You made my day 😁😆
To be honest I'd say that is too much retouching. The face is passable but the body is far too much and looks incredibly fake. Granted your final step was just a quick example but still.
Furthermore how are you going to retouch anything larger than 1024 for say print purposes since you can't upscale the photograph in SD.
Perhaps I'm missing something here but there are a ton of better Photoshop specific tools for exactly this purpose.
Not bashing SD, rather saying use the right tool for the job.
Hi Marchant, to your first point, yes I kind of rushed through the body for this example but you are able to take your time and generate multiple versions then mix and match the best generations to get more detail or less detail depending on what you’re after.
As for the 1024, when you’ve upscaled the image and set the inpainting to masked only it fits the 1024 pixels into what ever your masking. So keep the masks at roughly that size. If your upscaled image is 4096/4k don’t make the masks more that a 4th the size of the image. I hope that makes sense. As in do small chunks at a time. That way you retain the image quality.
I agree there’s other ways to do this using photoshop. This is just another tool in your arsenal to get the job done.
it would be nice not having to write prompt for every image