Fundamentals of AI Art Generation

[kakom]

That's honestly a major improvement over my experience trying to get text, that second one is almost correct!

What did you put into the prompt @kakom to make the text so readable? Every time I've tried, the text always has major typos or is illegible nonsense.

Great to see you in this thread @kakom . You have some excellent results here. Will you use them in your AAR?

Also, I have the same question as @jak7139 : how did you get the AI to give you a readable sign?

This was my first time using AI to create images. As for the phrases used, I just typed: Create an image of the golden raised text "Taurica Part III: Age of Thought and Discovery (1300 - 1800)". The first images I got were rather poor, so I started typing corrections like this word is incorrect, then I deleted the part 1300 - 1800, and after about 4-5 attempts I got more or less these two satisfactory results.

When it comes to the use of AI art, I wanted to use it in limited extent in part 3 of the megacampaign. I can create four to five images per day in ChatGPT in the free version, I don't really want to spend money on it so for now it's just an idea.

 

[CBR JGWRR]

This was my first time using AI to create images. As for the phrases used, I just typed: Create an image of the golden raised text "Taurica Part III: Age of Thought and Discovery (1300 - 1800)". The first images I got were rather poor, so I started typing corrections like this word is incorrect, then I deleted the part 1300 - 1800, and after about 4-5 attempts I got more or less these two satisfactory results.

When it comes to the use of AI art, I wanted to use it in limited extent in part 3 of the megacampaign. I can create four to five images per day in ChatGPT in the free version, I don't really want to spend money on it so for now it's just an idea.

Well, that's interesting. Just tried a batch of 8 in LeonardoAI with the same prompt, and only two were rejects for typos.

Of the six, I think this was my favourite:


Actually quite surprised at how well it turns out.

I then thought to expand the prompt by asking for it on a brown leather hardback book, and well...


In terms of overall layout and feel of a book on a table, this was the best, just a shame about the oddly arrayed "Tauriico" at the top...

Just done two more batches and it's given me pretty much exactly what I had in mind:


It's impressive how much better at text it's gotten. Last time I tried, it wouldn't be able to do this.

 

[Bullfilter]

And I thought I'd follow up on the iterative image regeneration I mentioned earlier. This iteration used Leonardo Phoenix 1.0.

First, I'd asked for a Landtag building with a large Frisian flag. I thought this was the best of the 4 images generated first up:

But the flag was wrong, so I then took this 'stock' flag I have (left) and generated an AI flag from that (right):

Then removed the background from the generated flag, pasted it onto the first image to get this:

Using that as a new reference image, generated this as the final version to 'tidy up the edges' for the final image:

Might have been one more step than necessary in there (the flag regeneration) but wanted the 'looks' to be as similar as possible.

The text prompt for the building and flag images stayed the same throughout.

 

[Rensslaer]

I chuckled when I saw this today on a spoof news site and figured you'd enjoy it also...



Rensslaer

 

[kakom]

So I was checking a few things to see how well the free AI CHatGPT image generation model would handle creating images based on other images. As part of creating images for AAR I thought I would use mosaics in the introduction to part 3. So I asked it to create a mosaic based on the image below and the results really surprised me.

Output image:

Result:

Leonardo AI free version completely let me down in this regard. Attempts to create a mosaic depicting a white griffin came out tragically.

 

[Chac1]

I must say I am very happy to see folks interacting in this thread. There are some recent additions here that are beyond expectations.

Well, that's interesting. Just tried a batch of 8 in LeonardoAI with the same prompt, and only two were rejects for typos.

First, I have to call out the efforts of @CBR JGWRR here. Here he is volunteering images and time to make images for another member of the community. That is truly extraordinary and commendable.

And I thought I'd follow up on the iterative image regeneration I mentioned earlier. This iteration used Leonardo Phoenix 1.0.

Also, thanks to @Bullfilter for this great explanation of "Iterative image regeneration." Your explanation and examples are a thousand times better than my attempt at an explanation.

So I was checking a few things to see how well the free AI CHatGPT image generation model would handle creating images based on other images. As part of creating images for AAR I thought I would use mosaics in the introduction to part 3. So I asked it to create a mosaic based on the image below and the results really surprised me.

Great to see these experiments and comparisons from @kakom . It appears paying for ChatGPT might be worth it if it can best Leonardo. (And here I was having issues with Microsoft Copilot today with it saying I had reached a 24 hour limit and I hadn't opened it in a week. Found a work around but such glitches and performance make you want to find alternatives. Glad to have an example of one such from kakom.)

I chuckled when I saw this today on a spoof news site and figured you'd enjoy it also...

@Rensslaer your jokes are always appreciated. At least I appreciated that one.

 

[CBR JGWRR]

Well... Happy to do requests provided I don't use all the tokens, which shouldn't be an issue.

 

[Sanvone]

Anyone has any experience with recent (2 days old) "Consistent Characters" feature introduced in many AI platforms? Cause I've seen it today and remembered that some people were complaining about not being able to replicate their characters between generations. From the tutorial it seems pretty straightforward.

P.S - Ah yes. Joined the AI generation wagon. Working on Leonardo.AI, OpenArt.AI and sometimes friend generates some things for comparison on ChatGPT.

 

[Chac1]

Thanks for the news @Sanvone . It is obvious from this thread that the AI realm is quite dynamic, changing all the time.

I have not been on Leonardo in several weeks. I have noticed @Bullfilter using a new Phoenix platform that Leonardo is offering. So is this new "consistent characters" feature offered on Leonardo? Or elsewhere?

Yes, some of us (perhaps led by me) have complained about this issue earlier in the thread (with examples of misses). Would be good to learn where we can get this problem solved. Thanks.

 

[CBR JGWRR]

So I was checking a few things to see how well the free AI CHatGPT image generation model would handle creating images based on other images. As part of creating images for AAR I thought I would use mosaics in the introduction to part 3. So I asked it to create a mosaic based on the image below and the results really surprised me.

Output image:

View attachment 1248421

Result:

View attachment 1248423


Leonardo AI free version completely let me down in this regard. Attempts to create a mosaic depicting a white griffin came out tragically.

View attachment 1248433

View attachment 1248432

View attachment 1248431

View attachment 1248434

Yeah, point goes to ChatGPT on this one. Best ones I've had trying the same here in Leonardo are these:



Most outputs I've had are horribly mutilated heads and spiral wings...



On LeonardoAI, the consistent characters (named "Character Reference") thing is only enabled on some models and presets, and isn't available yet for Phoenix 0.9 or Phoenix 1.0.

 

[kakom]

Yeah, point goes to ChatGPT on this one. Best ones I've had trying the same here in Leonardo are these:

View attachment 1248849View attachment 1248850

Most outputs I've had are horribly mutilated heads and spiral wings...



On LeonardoAI, the consistent characters (named "Character Reference") thing is only enabled on some models and presets, and isn't available yet for Phoenix 0.9 or Phoenix 1.0.

@CBR JGWRR these griffins resemble eagles


leonardo.ai is better at creating frescoes than mosaics but the number of tragic results is still too high for fairly decent images to get this one below I had to generate the prompt four times.

As for me chatgpt is a bit better only the disadvantage is the possibility of generating 3-4 images per day in the free version.

Now I have a question, what other generators do you recommend because I'm wondering whether BING from Microsoft is worth trying ?

 

[CBR JGWRR]

@CBR JGWRR these griffins resemble eagles

Yeah... They do. Best ones here doesn't necessarily mean they are actually good...

This is representative of what I got otherwise:

 

[Chac1]

Now I have a question, what other generators do you recommend because I'm wondering whether BING from Microsoft is worth trying ?

Well, @kakom you may want to go back to this earlier posting in the thread for further discussion on that score.

I am missing Playground AI's old iteration a lot these days. I find myself mostly using Microsoft Copilot. However, when the free version of Copilot annoys me (as it has recently) I fall back to Leonardo or Image FX. Those are my personal choices right now, but there may be other platforms that are useful too.

 

[Chac1]

A few updates for folks who follow this thread....

First, you can see some of the art that @kakom has created in the third installment of the mega-campaign tracking Taurica: Taurica Part III: Age of Thought and Discovery.

Also, there was this from @Sanvone ....

Anyone has any experience with recent (2 days old) "Consistent Characters" feature introduced in many AI platforms? Cause I've seen it today and remembered that some people were complaining about not being able to replicate their characters between generations. From the tutorial it seems pretty straightforward.

So perhaps other platforms besides Leonardo.AI have tutorials posted on "consistent characters." Yes, I found the Leonardo tutorial easy to understand and I wanted to use it, but discovered the feature is not available until an unspecified later date. Perhaps if that works the way it should I will then decide to pay for a platform.

But as an experiment, I decided to generate some art with Leonardo hoping it would get closer to providing a "consistent character."

To those who might be following my new spin-off AAR, Lost Myths, you may be familiar with this avatar for Chieftess Alvör:

​

This was an image created strictly through word prompt in the now defunct parts of Playground AI. This was my reference image in my experiments with Leonardo. The fact that this image came from a different platform is immediately going to cause issues with its replication as Leonardo warns in its "consistent character" tutorial. Still this was an experiment. Consistent characters works better with photorealistic subjects. For that reason I asked for Leonardo to create output that was photorealistic. The best of the four outputs I used in the latest post even though it was far from perfect.

​

In the post, I referenced this as the Disneyfied version of the chieftess. I had tried to place her at a feast and the AI did do that, so that was a welcome outcome. However, the chieftess seemed to have her hard edges sanded away and that wasn't for the best, in my opinion. That post also includes a photorealistic profile of the chieftess that I used with some prudent cropping. However, none were truly "character consistent."

Although it was too late to include in my posting for this week, last night while creating some other pieces, I returned to this output and tried a much more detailed verbal description. Judge what you think about the below image, but the lesson for me was to try building a more detailed verbal accompaniment to the reference image, and you may have better results. However, this is still not character consistent.

​

If folks know of AI platforms offering true "consistent character" features, please share if you will and that will certainly help some of us. I will definitely let folks know when Leonardo opens up that feature and I will also test it as I've done with other platforms.

 

[Bullfilter]

the lesson for me was to try building a more detailed verbal accompaniment to the reference image, and you may have better results

Yes, I definitely find the same. It still doesn’t always work even after multiple iterations, but it does improve outcomes and sometimes works really well. I’ve been able to regularly get roughly compatible character photos, but wouldn’t yet call it consistent.

 

[MacGowan]

Nice to see a solid AI thread about this stuff. I saw some questions about the various sites pros and cons. I figured I’d throw my hat in the ring.
I’ve been worldbuilding this little project i wanted for an Aurora+/Stellaris/Darkest Hour AAR (not sure how it'll work.) But here’s some stuff I’ve made for the lore:

So my current workflow is: farming images via Leonardo, MidJourney, CivitAI, Ideogram, Bing DALLE. I usually run a few a day. sort them by factions and awesomeness.

Then when I wanna make something, i pick an image. maybe gifaceswap the face if i need a specific person. Inpaint in MJ or Ideogram. Krita Interstice. that’s where I do the polishing of the inpainting, or make a nice img2img with low strength to make it all come together

Then photoshop the text and design. (though Ideogram has a good text ai inpainter)

 

[Chac1]

Thanks for your insights @MacGowan . Your insights have the ring of an expert. Please feel free to share some of your ideas and techniques if you feel the urge as the intent here is to help those who are just starting to deal with graphics for AARs.

A few questions, if you don't mind....

First, lovely graphics you have already assembled for your upcoming AAR. Will you keep us posted in this thread when you launch?

By "farming images" do you mean making new batches? Or collecting images from elsewhere?

What is your experience like with MidJourney? Are you using a free version? As you indicate you do face editing and other work in MidJourney, does it have an easy to use editor for such face editing ("swapping")?

I have used Krita in the past and it is a good platform although I prefer if possible to use editing platforms that are part of the image generator, just for ease of use.

What can you tell us about CivitAI and Ideogram? Are they similar to MidJourney or Leonardo?

How much Photoshop use contributes to the pictures you included? Is that a last polishing step for you?

Thanks for providing any further insights into your process.

 

[MacGowan]

Thanks for your insights @MacGowan . Your insights have the ring of an expert. Please feel free to share some of your ideas and techniques if you feel the urge as the intent here is to help those who are just starting to deal with graphics for AARs.

Glad it's useful! Not an expert by any means, more a graphic designer with an interest in AI. so my help would be more in that style.

First, lovely graphics you have already assembled for your upcoming AAR. Will you keep us posted in this thread when you launch?

Thank you, i appreciate that! Can do. I think it will take a little while tho. I'm trying to figure out what game to AAR, I need to be able to customise my solar system, have moving orbits, ground fighting, and a solid HOI military system. if someone knows such a game.

By "farming images" do you mean making new batches? Or collecting images from elsewhere?

making new images. Generating an image of a specific thing is HARD. So I have made a library of images i can pull from instead.

What is your experience like with MidJourney? Are you using a free version? As you indicate you do face editing and other work in MidJourney, does it have an easy to use editor for such face editing ("swapping")?

I don't think MJ has a free version. rn MJ looks the coolest, but has lower prompt understanding and coherency (prompt understanding meaning "white ball, on red square, on green triangle".) (coherency being how many fingers on one hand, etc)

for faceswapping i've been using aifaceswap. Not very good, but free, no login, and easy to use. (then some photoshopping afterwards.)

I'd recommend getting a month with MJ, to see if it's your thing. Leo, Ideogram and Civit are freemium.

I have used Krita in the past and it is a good platform although I prefer if possible to use editing platforms that are part of the image generator, just for ease of use.

I feel ya. my workflow is a bit cumbersome now, and moving from one platform to another compounds time for sure. however Krita w interstice is the easiest non-censored inpainter out there (some AIs block swords, guns, battles, etc).

What can you tell us about CivitAI and Ideogram? Are they similar to MidJourney or Leonardo?

- Civit is a bit of a buggy cesspool. But it does have the biggest model database for you to run. (SDXL, Flux, SD3.5, etc. or smaller custom stuff like "comic book style from the 60s")
- Ideogram is by far the most coherent and prompt understanding image AI. the style can be underwhelming, but if you want superman vs batman ontop of a green ball ontop of the red text "Hello" it is the only image AI I know that will do that
- Leonardo and MJ look the best out of the box. rn those are the ones i use the most.

How much Photoshop use contributes to the pictures you included? Is that a last polishing step for you?

this image is straight out of leonardo.

I use old version PS alot. even if it doesn't really need it, I like to make the images "mine." I do have a friend who hates photoshop and inpainting. He's a fan of MJ, and the simple MJ inpainter they got.

Thanks for providing any further insights into your process.

No problem! would love to see some more workflows here. I know the big thing now is ComfyUI. It looks really good, but seems like a f£$@!ng pain to set up.

 

[CBR JGWRR]

Honestly

Thank you, i appreciate that! Can do. I think it will take a little while tho. I'm trying to figure out what game to AAR, I need to be able to customise my solar system, have moving orbits, ground fighting, and a solid HOI military system. if someone knows such a game.

I'm not sure there is...

Star system customisation with orbits is tricky - very few games actually do realistic orbits as it is, and the only one I know that does custom star systems as well is Children of a Dead Earth. But CoaDE has no ground combat whatsoever, plus while you can design your own equipment (subject to the constraint it's limited to stuff there's hard numbers for known already) there's no kind of industrial requirment and logistics are abstracted except for certain missions where bringing refuellers with you is necessary.

I'd love a CoaDE 2 that includes these factors, but... the Dev is busy on a different project now.

 

[MacGowan]

Honestly

I'm not sure there is...

Star system customisation with orbits is tricky - very few games actually do realistic orbits as it is, and the only one I know that does custom star systems as well is Children of a Dead Earth. But CoaDE has no ground combat whatsoever, plus while you can design your own equipment (subject to the constraint it's limited to stuff there's hard numbers for known already) there's no kind of industrial requirment and logistics are abstracted except for certain missions where bringing refuellers with you is necessary.

I'd love a CoaDE 2 that includes these factors, but... the Dev is busy on a different project now.

My impression is that Aurora 4x has orbits? and Falling Frontier will have it, so i'm looking into those. I don't need full orbital mechanics, just want the planets to move. CoaDE looks like fun tho!

If nothing ends up fitting the bill. i might just build it manually; tabletop and spreadsheet style.