JD's bazaar

[doe.1971]

1. Start with run_anime.bat instead of run_realistic.bat Not a world of difference but it will load a model base called animapencilxl that's better suited for anime or drawing plus will change a few negative prompts from the realistic mode.
2. After loading check INPUT IMAGE and ADVANCED.

ex1.jpg (11.64 KiB) Viewed 4782 times

3. Go to INPAINT OR OUTPAINT tab
4. Into the lower box load your starting image.
5. Get rid of the text first, it ussually mess things up. Just select and run GENERATE with nothing as prompt.

6. Choose the one that removed the text best from the upper window.

[doe.1971]

6. To move the image without text to be further processed, drag and drop it (in Chrome) from the upper window into the lower window. Notice that the selection tool is still active.

7. Clear the selection by hitting the clear icon in the upper right corner of the lower window.

ex5.jpg (3.17 KiB) Viewed 4782 times

8. You're ready for the next iteration and your image look like this in the lower window. Notice there's no selection.

[doe.1971]

9. Select the upper clothing.

10. GENERATE with the favorite prompts. I'll use naked as a prompt here.
11. From the offered choices, I pick this one.

12. Drag and drop it from the upper window into the lower window like before.
13. Clear the selection.
14. Select the panties.

15. With the same nude as prompt do another GENERATE

[doe.1971]

16. I prefer this one.

17. Now you have a chain going nowhere. You can select it and removed it with no words in the prompt. I got this.

18. Being a time piece, I wanted some pubic hair so I marked the area, used pubic hair as prompt. I chose this.

[doe.1971]

19. I wanted to get rid of all the distracting text, so I select it and GENERATE.

20. I wanted to expand to the right a bit, to make the burly guy there more visible, so go to the OUTPAINT direction I checked RIGHT and GENERATE, with nothing as prompt, as I rather let the algorithm interpret the image than me forcing into something specific. I got this.

21. If you want a more wholesome image you can expand toward the bottom, so for this step select BOTTOM instead of RIGHT. I used naked, bruises, cuts, standing as prompts here.
22. Be careful here as there's a black band on the bottom of the image that will produce a black strip plus AI will be forced to figure out an object there.

23. Since there's no crop function on Fooocus you gotta COPY the image then past it in PhotoShop, remove the lower band save and then bring it back again in Fooocus.

[doe.1971]

Finally, I got this and called it a day, it's already a 8000x5000 pixels file.

[doe.1971]

You probably want to remove that artifact like water on the lower right, just select it with nothing in prompt, generate and choose a result that has it removed. I'll hit the sack, myself.

[doe.1971]

One day, when Radovid visited Emhyr, in Vizima.

"While it's true that demeritium rings around their necks or torsos work well in supressing their magic abilities, I found that a rod of demeritium inserted in their asses and then tightened around their necks is far more effective, in terms on esthetics and personal discomfort" explained Emhyr.

"I should have my blacksmiths made some too, then'. answered Radovid "My witchhunters still capture warlocks and sorceresses almost every week. It will be a long time before Redania is free of their scourge. We don't wanted to be too cozy in captivity, miserable connivers".

AI didn't improve anything here (at least according to my questionable taste) so they are virgin on this aspect.

[bleumune]

1. Start with run_anime.bat instead of run_realistic.bat Not a world of difference but it will load a model base called animapencilxl that's better suited for anime or drawing plus will change a few negative prompts from the realistic mode.
<snip>

Thanks a lot for the help! How'd you figure all this stuff out?

Here's what I came up with. I had a couple of things to figure out. I had to save intermediate processed images to my hard drive since the drag-and-drop doesn't seem to work with Firefox (I refuse to use Chrome since who knows what Google might look for, since they own Chrome).

A couple of things:
- When I ran run_anime.bat, I didn't get ANY negative prompts. It still seemed okay with the prompt "nude" but I wondered.
- Also, do I need an anime file like sdxxl_v3.0.safetensors? The default load is PencilXL.v310.safetensors.
- I had real difficulty trying to use AI on the face. Nothing was realistic. That's when I started to wonder if I needed some negative prompts, but that'll need some research (unless you've got the answer right at your fingertips as usual!)

The way you used Outpaint to expand the drawing was really interesting! I'll have to try that on some of my favorite Men's magazine covers.

[doe.1971]

Thanks a lot for the help! How'd you figure all this stuff out?

Glad to help. Played a few days rather intensively with my fav performers.

- Also, do I need an anime file like sdxxl_v3.0.safetensors? The default load is PencilXL.v310.safetensors.

There's a constant maelstrom of new base models coming out every day. I'm sure there will be better ones soon. Keep an eye on https://civitai.com/. But even the default one gives me decent returns.

- I had real difficulty trying to use AI on the face. Nothing was realistic. That's when I started to wonder if I needed some negative prompts, but that'll need some research (unless you've got the answer right at your fingertips as usual!)

I tend to leave faces unchanged, I like the person behind the image, I don't want it replaced with an attractive nobody.

The way you used Outpaint to expand the drawing was really interesting! I'll have to try that on some of my favorite Men's magazine covers.

Increase the output to minimum 10 instead of 2, the amount of error you can get when OUTPUTTING is rather large. Sometimes I gotta run multiple batches of 10 to get smth I could keep.