First Splinterlands Art Contest
Hello #Splinterlands art community members! This is my first entry in the contest. It was recently suggested I should give it a try. Tbh I never knew when it was happening. Now that has been fixed. Not everyone knows everything about HIVE and some of us (like me) need our hands held apparently 😅 Anway, so here it goes... first entry. Alt version of Heloise the Hollow.
Why Heloise?
First. The easiest question. Why Heloise? Umm... well... because she's the hottest, most talked-about, most-powerful, most sought-after card in the entire game right now. That's basically why. What other card could I start with. I mean, the kids these days, am I right? They like the fresh cards. And the fresh style 🤷♂ What can you do.
How did you do that?
Second, I understand I need to explain how I made this? Oof. That's just, wow. That's about as personal as anyone can ever get with me. Why don't you ask me for a naked pic instead, that would be less invasive 😂 I mean srsly. So intimate. Anyways, it's part of the rules and it's (I guess?) part of being in this community, so... here it goes... getting naked.
Photoshop Firefly Generative AI
Okay, so this image above---and basically EVERYTHING I do lately---is generated using photoshop with many, many assists from its new firefly generative AI tools. If you've noticed a substantial improvement in my art over the past year, that is most definitely why.
I even use the beta test versions. Especially lately. Why? Because they're releasing new firefly updates all the time. For a while there, the pace felt pretty ferocious. A lot of that doesn't get to regular Photoshop right away, and some of it never gets there. That's why I always try all the newest beta test versions. I literally just updated to the latest right before making the Heloise alt art above. Luckily it worked well and didn't short circuit the process I've been using.
Starting the Image
First, I started with a new image file. Blank white screen. Yes exciting stuff. I tend to start with 4x4 and 720 pixels/inch. It doesn't really matter. But that's my default to start. Here's a screenshot below:
I started with this. A new image file. 4x4 at 720 pixels/inch.
Firefly Prompting
Next, I used firefly prompting and importantly, DID NOT input a reference image. I'll tell you right there, that's important. It sounds counter-intuitive, but it's the way I've found to keep things high-quality and consistent. Use your words instead of someone else's picture. Describe what you want to see in painstaking detail. Like at least a paragraph describing every key aspect of the image. It's almost like writing a story at that point.
How well can you describe the scene you see in your head, in words? I'm not giving you my words though, sorry that's too personal. If you mess around with firefly for a year and finally figure out the right way to talk to her, so she plays nice with you 😂 then we'll info-share on the technical nitty-gritty of prompting secrets.
One important point here. DO NOT input a reference image. It will look very stiff. Metallic and life-less. At least with firefly. Instead, get to know the generative prompts that firefly likes. It took me a long time to get it. I'm just finally hitting a groove. Firefly is an AI entity, take it slow and treat her kindly. You'll end up with some great stuff too.
3 Iterations Indefinitely
Unless you change this feature, firefly will generate 3 iterations of every prompt you feed it. Indefinitely. 3 more. 3 more. 3 more. Until it literally freezes and you have go edit>purge>all to make her be your friend again. That's another important part. You almost definitely will NOT get what you want on the 1st try. Or the 2nd. Or the 3rd. For the Heloise above, I actually got what I was looking for on the 5th try so after about 15 iterations. That's pretty good. Screenshot below.
After generating about 15 iterations, I knew this was the one instantly. The things I looked for were visual impact factor and consistency with my prompted style and vision.
Time to Clean
Next it was time to clean some stuff up. If you've used generative AI for image generation before, you know that sometimes (a lot of times!) it does some pretty weird stuff. You gotta be able to quickly, efficiently and easily smooth that out and bring it in line with the image. Tbh, that's why I love photoshop. It just makes the cleaning and polishing so much faster and easier. Plus, it's all still powered through firefly prompting.
For example, in the screenshot below I zoomed in on Heloise's face. You can see it doesn't look great. I actually know what firefly was trying to do. I was asking a lot. It just didn't quite get there.
This kind of junk is common in generative AI art. It needs to be cleaned and polished to make it look good and be consistent with the rest of the image.
In this case, I lasso those areas specifically and run firefly over it again, prompting again exactly what I want there. New prompts. Focused only on the face. It got it right after about twenty or some plus iterations. Screenshot below:
One reason it's best to start with a relatively high pixel ratio is so you can zoom and fix stuff like this.
As just another example, check out the next screenshot below. What is that? I think I know what firefly was trying to do. Again though, fail. It definitely still makes a lot of little fails that need to be corrected---even in its best images.
What is this? A rip in the fabric of space and time? That's cool. Just not what I was going for.
Again, firefly makes it efficient to fix. Lasso and tell it what you wanted to see there instead. Use good descriptive words. Describe that section in detail. Screenshot below:
Here the fix was easy because I mostly just wanted that strange "rip in the fabric of space and time" to be removed.
Added More Depth
Next I zoomed out to give the scene more depth...
Expanding the background.
Filled the empty blank canvas with new prompts...
Adding new prompts for the expanded background.
After about 6 iterations, settled on a darker background with blue fade at the bottom...
The expanded background darkened the scene.
What's the Different Between These 2 Pictures! 🔎
After that it was just fixing and fixing. Polishing it all up. There are probably 100 little details cleaned and polished between the image above and the one below. 99% of the time, polished using prompting. In the end, it looked like this...
Lots of subtle changes were made. If you like playing those games "What's different between these two pictures?" then here you go 😂 You will love this. Point them all out in the comments! Compare this image to the previous one.
Added Text
I also wanted to add text with the name of the character. But didn't want it to look too stiff either. Prompting from a new image file, I ended up getting this mish-mash...
Firefly is good at lots of stuff, but spelling is not one of them. And this was the BEST of many, many, many iterations 🤷♂ It just can't spell.
Btw this is 100%, all the time, how firefly is. I don't actually know how other platforms handle "spelling" but it's one of the few areas where firefly can never quite get it right. What does that mean? Lasso, expand, prompt it some more. Eventually I got it to look like this...
Getting text right is one of the most painstaking parts of the generative approach. Right now, you literally have to hold its hand. I bet that will improve eventually.
After that, it was selecting around the outline. Used a combination of paint brush and more prompting to get a clean outline around the text...
Final text, cleaned and polished.
Selected. Inversed selection. Copied and brought it over to the Heloise image. Usually I'd merge layers right away to save on memory, unless there's a reason not to. Here there is a reason. We'll get to that shortly. Anyway, so I left the text as a new layer...
Also expanded the background a bit more, to accommodate the text.
After that, used prompts again to expand the background...
The new blank canvas was filled with a basic background prompt.
I liked the idea of a vertical image, similar to how SPL card are, so just refitted the frame...
For a vertical card-like image, I just quickly reframed the image. That new white canvas at the top was filled with another simple background prompt.
Added Transparency and Adjusted Color
Like I mentioned above, I kept the text layer separate on purpose. That was to give it a transparency filter. Here I chose pin light.
Of all the transparency filters in photoshop, I tend to almost always use hard, vivid, linear or pin. This time I chose pin.
NOW I could merge the layers and throw on a last minute adjustment filter. It wasn't needed but why not? I love to make the colors pop. There are many awesome adjustment filters these days, but I often go with one literally called "color pop" 😅 Here for Heloise, I stuck with my gut and chose color pop again...
I like almost every adjustment filter in photoshop. It just depends on what you're doing. For stuff like this though, I love the color pop adjustment that intensifies and more fully saturates the neon colors.
Conclusion
And... that's it. Once you have the hang of it, it can be a very productive way to generate a lot of great images in a short time. Plus, look at this picture 😍 It produces some consistent, high-quality stuff. Hope everyone enjoys. If you like this, think about following me. My posts usually have at least 4 to 5 images I make like this. Often times animated. Oh geez, now I guess I need to show you all how to make super-easy, super-fun GIFs with firefly. Next time!
What about Survival Prep?! Btw if you're looking for my 7-day series on Survival, I'll release the next installment on earth splinter tomorrow. Art contest today!