Writing & Publishing Getting a Book Cover

[BKHunter]

Why do I have the feeling you have spent as much time coming up with a book cover as you did writing the book?

But this is interesting.

Click to shrink...

Thanks for the replies! The book took me years to write; this is just a few minutes here and there when I have time to set everything up, then a few more to check the hundred or so images later. It did take a few hours to load A1111, though. DiffusionBee was an easy installation if anyone wants to try that route.

To be honest, this is a lot of fun and I have had some success with photography as well, including publication, exhibitions, etc., so I can't resist this type of thing. If it's any consolation, the next book cover should go a lot faster. Btw, if anyone else is thinking of trying this, I'd recommend 32GB of memory -- I am having some issues with only 16GB and an M1 iMac.

I'm not done playing with A1111 yet, but here's a sample recent image. I really should post some of the bad ones for laughs, as well.

G

 

[carolinamtne]

I know you've been wanting to include the crescent moon, but the Arab garb on the man implies the same thing. (If I remember correctly, the name of your book is The Crescent Moon, or something like that, so it makes sense.)

 

[M. D. Boncher]

rab garb on the man implies the s

Click to shrink...

Btw, if anyone else is thinking of trying this, I'd recommend 32GB of memory -- I am having some issues with only 16GB and an M1 iMac.

Click to shrink...

Your issue is probably more your graphics card. I just upgraded from 6gb to a 12gb card. night and day difference. Also, with the exception of whatever Mac does, it's gotta be an NVDIA graphics card. The software is specifically designed for that. Sorry, Radeon users. The support isn't there yet.

 

[BKHunter]

Your issue is probably more your graphics card. I just upgraded from 6gb to a 12gb card. night and day difference. Also, with the exception of whatever Mac does, it's gotta be an NVDIA graphics card. The software is specifically designed for that. Sorry, Radeon users. The support isn't there yet.

Click to shrink...

No can do on an iMac -- it has integrated graphics and you can't add a graphics card. By the way, there is supposed to be an Apple "Core ML" graphics update out with enhancements that will half the Stable Diffusion image generation time, but it's apparently still not quite ready for prime time. If anyone wants to try it out, Guernika is a cheap Mac app that implements it (512 x 512 max resolution, though).

 

[carolinamtne]

At least you're having fun.

 

[Zee]

Thanks for the replies! The book took me years to write; this is just a few minutes here and there when I have time to set everything up, then a few more to check the hundred or so images later. It did take a few hours to load A1111, though. DiffusionBee was an easy installation if anyone wants to try that route.

To be honest, this is a lot of fun and I have had some success with photography as well, including publication, exhibitions, etc., so I can't resist this type of thing. If it's any consolation, the next book cover should go a lot faster. Btw, if anyone else is thinking of trying this, I'd recommend 32GB of memory -- I am having some issues with only 16GB and an M1 iMac.

I'm not done playing with A1111 yet, but here's a sample recent image. I really should post some of the bad ones for laughs, as well.

GView attachment 13835

Click to shrink...

Dude’s like….”Whoah….”

 

[BKHunter]

Dude’s like….”Whoah….”

Click to shrink...

Might add some small oil derricks in line with the Middle Eastern theme and the book; what do you think? I'm currently running this version through multiple iterations to learn inpainting, btw. I got rid of the overly-processed clouds and added more heat lightning already.

 

[M. D. Boncher]

The pieces of the image are still not integrating well. Are you using an artist or art style in your prompt? That would help it a lot. Otherwise you get clipart collage in most cases. Okay, *I* get clipart collage when I don't start specifying that stuff.

 

[BKHunter]

Still fooling with this (no rush since I can't publish for another half year or so) and had a style question. Which do you prefer: big moon or small? Note that the name of the novel is Under the Crescent Moon.

 

[carolinamtne]

Hmmm. The guy with the small moon only has three fingers on each hand. The back of the head with the big moon looks weird.

But I think I like the smaller moon better. The big moon looks too big to be realistic.

 

[BKHunter]

The pieces of the image are still not integrating well. Are you using an artist or art style in your prompt? That would help it a lot. Otherwise you get clipart collage in most cases. Okay, *I* get clipart collage when I don't start specifying that stuff.

Click to shrink...

Yes to both — but I’m using multiple artists so it (should) be less derivative.

Hmmm. The guy with the small moon only has three fingers on each hand. The back of the head with the big moon looks weird.

But I think I like the smaller moon better. The big moon looks too big to be realistic.

Click to shrink...

Stable Diffusion is infamous for messing hands up; you would have to redraw them until they're right. Thanks!

 

[M. D. Boncher]

I'm just gonna throw in a word of caution if you are looking at printing through Ingram Spark. Get their cover template first before getting too deep. The typography is killing me right now as I misaligned it due to the bleed areas are ambiguous. Not to mention out of gamut color warnings, and then discovering because of how I did the titling, I have artifacts I can't get rid of.

Oh... and it's gotta be done asap so I can order as soon as possible. This is why you want to hire an artist if you can.

...my brain hurts.

 

[BKHunter]

I'm just gonna throw in a word of caution if you are looking at printing through Ingram Spark. Get their cover template first before getting too deep. The typography is killing me right now as I misaligned it due to the bleed areas are ambiguous. Not to mention out of gamut color warnings, and then discovering because of how I did the titling, I have artifacts I can't get rid of.

Oh... and it's gotta be done asap so I can order as soon as possible. This is why you want to hire an artist if you can.

...my brain hurts.

Click to shrink...

Thanks for the advice! I've got a while to go before that stage. I agree it would be best to hire an artist if possible. Eventually, I think they'll end up using disruptive software like Stable Diffusion to speed up their processes, but it requires too much tweaking to put them out of work.

By the way, what are you using for a negative prompt? Here's what I'm hearing is best for SD model 2.1:

Negative prompt: ugly, tiling, poorly drawn hands, poorly drawn feet, poorly drawn face, out of frame, extra limbs, disfigured, deformed, body out of frame, bad anatomy, watermark, signature, cut off

The long negative prompt does seem to make a difference, although perhaps not as much as one would hope. Here's an example with all of that enabled:

 

[M. D. Boncher]

Eventually, I think they'll end up using disruptive software like Stable Diffusion to speed up their processes, but it requires too much tweaking to put them out of work.

Click to shrink...

Exactly.

As for the negative prompts, I have usually up to 30 terms in 1.5. 2.1 is not ready for prime time as far as I'm concerned. That image looks a lot better.

BTW, for more AI art info here's my latest blog article on the subject with a couple new images from me for those who are interested.

We Were Forewarned This Was coming, & Other Thoughts

AI art continues to be a source of white hot contention in the art community. With the arrival of ChatGPT, even more fuel and oxidizers are being added to the fire. Corporations now see dollar sign…

resonantpoint.wordpress.com

 

[BKHunter]

Thanks for sharing the link; I like your "Trial of the Future" drawing. Are you making ones like that at 512 x 512 resolution and then upscaling them? One of the things that convinced me to try the 2.1 model was that it supports 768 x 768 reolution from the get-go, but I recently downloaded the 1.5 models as well after reading that 1.5 is better at drawing people.

As you mentioned in the link, AI software seems to have blown up during the past week or so, with both ChatGPT and AI drawing programs getting a lot of press, etc. The next few months should be entertaining.

 

[M. D. Boncher]

Thanks for sharing the link; I like your "Trial of the Future" drawing. Are you making ones like that at 512 x 512 resolution and then upscaling them? One of the things that convinced me to try the 2.1 model was that it supports 768 x 768 reolution from the get-go, but I recently downloaded the 1.5 models as well after reading that 1.5 is better at drawing people.

As you mentioned in the link, AI software seems to have blown up during the past week or so, with both ChatGPT and AI drawing programs getting a lot of press, etc. The next few months should be entertaining.

Click to shrink...

That's just a raw image using a new model that came out. blanking on the full name but it uses the trigger word "manchu". Give very retro-futurist results that I've been looking for. I think that image is like 448x704 since I got my bigger gpu I can do bigger sizes. Otherwise, yes, I'll upscale using GFPGAN or LANZCOS.

I'm strictly SD 1.5 right now.

The next few months are going to be a very mixed time truthfully. Lots of good and bad are about to happen. BTW, beware doing automatic updates for A1111 (or anything really. Been lots of cases where there's been flaws. Had a huge error happen with dreambooth, just when I was ready to start making my own decopunk model. But right now, I'm in the crunch to get my book formatted for the printer and then ordered so I can't do it... yet.

I'm just keeping an eye on the news and putting updates on my blog. So, if you want the updates, that's a good place to go.

 

[M. D. Boncher]

Just to give you an idea of how crazed it can be to set up a book cover yourself, here's a screen cap of what I've been doing. This is about 14 hours of work, just using the printer's template. All those blue lines are marks you need to make stuff line up, not get cut off and hopefully... hopefully... you won't botch something that the printer will be like "fix this and this and this"... I've done all this work for 5 errors. Some were relatively minor, but others had to be addressed. Since I want to get the best cover and layout I'm able to, tada! Here I am. (It didn't help to discover that my titles all had to be redone because they came in as outlines (graphics not text) and the FX I was using to make them as fancy pants as they are were triggering on artifacts.

So this is me, knowing somewhat what I'm doing. If I considered all the time I put in for this cover and all it's elements, I'd call it about 3 full 40 hour weeks over the last few months. If this is what you consider a good use of your time to get something CLOSE to a professional cover, I encourage you to develop your skills and go forth to pursue you're vision like an auteur director. It can be lots of fun, but when you're trying to figure out why two letters won't line up properly at 2am because something broke in your illustration... You question your life choices.

But this is the best I can do with the tools I have and the time I've overspent. I've gotta send this to press so I have books in time to sell by hand in 3 weeks now. And that will be a huge relief once they show up.

 

[BKHunter]

That's just a raw image using a new model that came out. blanking on the full name but it uses the trigger word "manchu". Give very retro-futurist results that I've been looking for. I think that image is like 448x704 since I got my bigger gpu I can do bigger sizes. Otherwise, yes, I'll upscale using GFPGAN or LANZCOS.

I'm strictly SD 1.5 right now.

The next few months are going to be a very mixed time truthfully. Lots of good and bad are about to happen. BTW, beware doing automatic updates for A1111 (or anything really. Been lots of cases where there's been flaws. Had a huge error happen with dreambooth, just when I was ready to start making my own decopunk model. But right now, I'm in the crunch to get my book formatted for the printer and then ordered so I can't do it... yet.

I'm just keeping an eye on the news and putting updates on my blog. So, if you want the updates, that's a good place to go.

Click to shrink...

I've been watching the news, as well. The AI graphics program lawsuit contains several inaccuracies (SD is a collage tool containing all Internet art in 5 GB? Really?). Still, it could go through if they get a gullible judge, so it bears watching.

It's not the first time I've seen something like this. First, computers were going to take our jobs. In reality, they just increased productivity, much as AI should ultimately do.

 

[BKHunter]

Just to give you an idea of how crazed it can be to set up a book cover yourself, here's a screen cap of what I've been doing. This is about 14 hours of work, just using the printer's template. All those blue lines are marks you need to make stuff line up, not get cut off and hopefully... hopefully... you won't botch something that the printer will be like "fix this and this and this"... I've done all this work for 5 errors. Some were relatively minor, but others had to be addressed. Since I want to get the best cover and layout I'm able to, tada! Here I am. (It didn't help to discover that my titles all had to be redone because they came in as outlines (graphics not text) and the FX I was using to make them as fancy pants as they are were triggering on artifacts.

So this is me, knowing somewhat what I'm doing. If I considered all the time I put in for this cover and all it's elements, I'd call it about 3 full 40 hour weeks over the last few months. If this is what you consider a good use of your time to get something CLOSE to a professional cover, I encourage you to develop your skills and go forth to pursue you're vision like an auteur director. It can be lots of fun, but when you're trying to figure out why two letters won't line up properly at 2am because something broke in your illustration... You question your life choices.

But this is the best I can do with the tools I have and the time I've overspent. I've gotta send this to press so I have books in time to sell by hand in 3 weeks now. And that will be a huge relief once they show up.

View attachment 13859

Click to shrink...

Thanks for the heads up; this could save me a lot of time later. Good luck!

 

[M. D. Boncher]

I've been watching the news, as well. The AI graphics program lawsuit contains several inaccuracies (SD is a collage tool containing all Internet art in 5 GB? Really?). Still, it could go through if they get a gullible judge, so it bears watching.

It's not the first time I've seen something like this. First, computers were going to take our jobs. In reality, they just increased productivity, much as AI should ultimately do.

Click to shrink...

It's why the plaintiffs are going to lose. Getty Images is going to make a play for it too, because their watermarked images have been scraped and their watermark is showing up in images. The funny thing is, I know it's showing up in pictures they don't own because of running a huge batch of "regulation images" for trying to do a dreambooth model. Because they own so many "editorial photos" the model thinks that it is typical to have that watermark on any image made to look like an "editorial photo" no matter the image. A sign of overtraining apparently