Men's Magazines alterations

[bleumune]

I just trained the untrainable LL, so for you it should be easy.

Maybe not. I must be doing something wrong since my attempts to generate a new face only give me a face that looks unchanged. I've tried 30 times and get the same result. Is there a difference between Inpaint and Outpaint?

I wonder about step 9. I used the cursor to circle the face. Is that correct? I used the prompt "screaming, tortured".

Thanks for your help!

bleumune

[doe.1971]

Step 9 marks the portion of the image to be replaced like this.

Inpaint replaced what you've selected (and now is in white). Outpaint expands an image (fill the extra space in the image with procedural generated stuff).

Click ADVANCED, goto MODEL tab on the right. See if you have a base model loaded. Should look like this. I used a nudity base but can be anything for heads.

In my case I replaced the body with a nude one, because I kept the head and replaced the body.

Also make sure you fill in completely the head place, not just the contour.

[bleumune]

Step 9 marks the portion of the image to be replaced like this.

That was my big error - I thought I needed to circle the area to be replaced (I think I got that from some webpage on AI that I reviewed.

Inpaint replaced what you've selected (and now is in white). Outpaint expands an image (fill the extra space in the image with procedural generated stuff).

I'll figure out Outpaint later!

I used a nudity base but can be anything for heads.

Where do I find a nudity base? I tried using the name of the base model you used, but nothing happened. Fooocus didn't automatically find that file.

Here's the best example I got from my efforts. Okay, but not up to what I've seen in other places. There's one thing for sure about this method of using AI - it works for one-off images, but trying to do a story isn't going to work, since the images will be randomly different.

Still, it was an educational process, but I have a lot to learn!

Thanks again,

bleumune

[doe.1971]

I'll figure out Outpaint later!

Outpaint is good when you want to extend an image vertically or horizontally. Does a pretty decent job interpolating the visual elements in your image.

Where do I find a nudity base? I tried using the name of the base model you used, but nothing happened. Fooocus didn't automatically find that file.

https://civitai.com search for SDXXXL

Here's the best example I got from my efforts. Okay, but not up to what I've seen in other places.

Still way better than the original.

There's one thing for sure about this method of using AI - it works for one-off images, but trying to do a story isn't going to work, since the images will be randomly different.

Not at all. AI uses a random seed (face in this case) unless it's instructed to stay with the same seed. Dig in the underbelly of Fooocus and find the Randomize Seed or like that, unmark it and it will stay with the same face, but arranged to the requirement of the prompt and the rest of the details in the image.

Still, it was an educational process, but I have a lot to learn!

We'll stop doing it when dead.

[bleumune]

Where do I find a nudity base? I tried using the name of the base model you used, but nothing happened. Fooocus didn't automatically find that file.

https://civitai.com search for SDXXXL

I tried that but I only got four hits, all for celebrities: Jenna Ortega, Jessica Alba, Kate Upton, and Alanah Pearce (whoever she is.) I even joined Civitai but got the same four hits.

Plus, where do I put the base model files when I find something, and what is this thing about Lora and checkpoints?

[doe.1971]

I tried that but I only got four hits, all for celebrities: Jenna Ortega, Jessica Alba, Kate Upton, and Alanah Pearce (whoever she is.) I even joined Civitai but got the same four hits.

Someone need updated prescription. https://civitai.com/models/167764/sdxxxl

Plus, where do I put the base model files when I find something, and what is this thing about Lora and checkpoints?

Checkpoints, different name for base models. Large files, some 6..7 GB each that tells the AI about a specialized field. There's generalist ones, like SD1.5, 2.0, 2.1 Jack of all trades and there are specialized ones, like SDXXXL (nudity) or others about Animes or classical painters. I prefer Juggernaut, the default for Fooocus, as it gives decent results.

LORA are much smaller files, that contain specific info about a certain celebrity or specific painter, in the previous example. LORA's are trained to supplement certain base models (or checkpoints, same thing) and work best in tandem with those base models.

When you download a certain LORA makes sure to notice what base model was trained for. If it says SD1.5 you're not in luck as this specific implementation, Fooocus doesn't work with SD1.5 , it has its own generalist base models.

You can use another SD distribution instead of Fooocus, that works with SD1.5 if you totally must run a certain LORA that only works in SD1.5 you can try Automatic1111 that's not as user friendly as Fooocus but accepts SD1.5

Also important in LORA is the trigger word, the word(s) you must include in the prompt to appeal the info in LORA. It's listed on civitai, when you download LORAs.

Finally, the place to load them is to click ADVANCED checkbox, then MODEL, and there you enter your base model and the LORA(s). You can use up to five LORAs simultaneously with various degree of strength but only one base model (or checkpoint).

The location is in the folder MODELS under the main folder instalation for Fooocus, subfolder CHECKPOINTS for basemodels and subfolder LORAS for ... LORAs.

[bleumune]

Shock Mystery Tales 10-62

This is one of my favorite Men's magazine covers from the '60's, but it suffers from the same malady of all those covers - the heroines are quite overdressed! I like this cover a lot because the heroine expressions are more accurate to the situation - horror and terror.

With a lot of help from JD, I was able to rectify the cover art to remove all the lettering as well as the clothing of two of the three heroines. The third one (in the background) was too fuzzy and distant - my AI program barfed when trying to remove her bra and panties. So, anyway, here's the best I've done so far.

[wjhk93]

Terrific redo in keeping with origin artist intent. The large breasts cup size with swollen areola tits is erotic great touch. You and I think alike regarding breast sizes. The bigger the better.
I personally would of replaced everything with poser characters, props and environment.
This is better for keeps with the true artwork. Will Burner

[frog]

Shock Mystery Tales 10-62

This is one of my favorite Men's magazine covers from the '60's, but it suffers from the same malady of all those covers - the heroines are quite overdressed! I like this cover a lot because the heroine expressions are more accurate to the situation - horror and terror.

With a lot of help from JD, I was able to rectify the cover art to remove all the lettering as well as the clothing of two of the three heroines. The third one (in the background) was too fuzzy and distant - my AI program barfed when trying to remove her bra and panties. So, anyway, here's the best I've done so far.

https://ralphus.net/thegimpforum/downlo ... &mode=view, the way MAM's covers should have been. Thanks.

[tan2000]

Shock Mystery Tales 10-62

With a lot of help from JD, I was able to rectify the cover art to remove all the lettering as well as the clothing of two of the three heroines. The third one (in the background) was too fuzzy and distant - my AI program barfed when trying to remove her bra and panties. So, anyway, here's the best I've done so far.

Awesome! Could you remake the other covers like this one? With AI seems the covers can have great enhancement for the alternations.