Improving a Client Project
by Replacing AI Art with Human Expertise
-
Undisclosed Project.
Leading an Art Team that Struggled with AI Images.
Here’s a study on how I helped a client improve their game art by replacing AI with Human expertise.
This studio had problems when trying to use AI for their concept art and game pitch. Every time they turned in their work, the dreaded "We need more changes" came in.
The art below was approved on the first try.
As you scroll down, you’ll find strategies and design techniques I have applied in this project, which you can mirror to improve your chances of winning over a board of directors.
Are you struggling with AI art solutions? Many projects fail because they rely too much on AI and bring in professionals too late.
Don't let your project flop or get canceled. Get top-notch concept art that meets your needs!
Get in touch here and make art work for you instead of trying to fit your project in the art AI gives you.
This was the first image I created. I sent it to the client to see if I was on the right track, and they approved it from the start.
They added it to the project guidelines and I was invited to lead the art team.
Many find that human ideas are better than AI, but AI is faster, and even though AI creates something more generic, it’s good enough for the market to buy.
I was eager to take on this project because what I found was the opposite: human wisdom, skill and intuition are important. The ability to tell right from wrong and truth from lie.
This was my first design, and I planned to make it better by adding more color and to get our players to react. But the client thought it was “good enough” as it was and suggested I focus on other important tasks instead, we could just send it to a Junior for polishing instead.
Artists are hired for their vision, not the software they have and their skills with a pencil.
How AI Art Sabotages Your Brainstorming
Many people asked me if you could still use Gen AI for the brainstorming phase and moodboards.
My answer is: Yes, you can!
Do I recommend it? Absolutely not!
Here's another image we had, generated by an AI artist.
Looks great, right? The lighting is good, the anatomy too, the detail is there... just... two questions:
How did he put the jacket on?
How is he going to take it off?
If you make the jacket functional, you will remove the sci-fi look, but keeping the sci-fi look will not make it functional. This is why I recommend starting from scratch and why good research matters, rather than coming up with something random and not referenced to reality.
I was already wondering about re-creating it with Marvelous Designer. The best tool I know for designing outfits, but we managed to make this one more usable and get it approved.
Research That Improves Projects and Gets Investors to the Yes
For some projects, I prefer to have a week or more to research. This might seem like a lot, but it helps us create a vision we don't have to wonder whether it is the right one for our project or not. The last thing we need is to settle on something early because we wanted it done fast, and then keep reworking it for the following months.
The budget will keep burning, no one will agree on a look, and by the end, players just scroll past the project, and no one knows why. The truth is, if you were never confident about your work, why would the players buy it?
To help our art teams work well, I plan check-ins for us to make sure we're on track. I like to plan early on, and check-in on progress by the middle of a project.
For example, when working the character we saw, I take what the client says they want to achieve, and turn it into a clear plan. Then, I set milestones to track whether we’re reaching our goal or not.
“Our Objective is: Design a character of a high ranking police officer. Visually conveying his dual role as protective, trustworthy parent and intimidating authority figure in a futuristic, tightly-controlled city.
Visual Checkpoint 1: Make the bases provided more functional and less like a costume.
VC2: Reintroduce sci-fi techwear elements later, once the function is established.
VC3: Research ACRNM and Vienna's 1990s academia movement for reference.
VC4: Make the silhouette more athletic, for authority and reliability.
VC5: Find a reference to tweak the body language for a calm but authoritarian energy.
VC6: Establish as many sharp, angular secondary reads as possible.”
This was an example check-in before we start sketching, this minimizes revisions and sets expectations. Sometimes it’s best to scribble something quickly over the image they have; sometimes it’s best done in writing. In this case, I did both.
Find out more about one of the many tools I use for setting goals.
References are important for our project, but we need real ones that match our goals, not just pretty pictures.
There’s a lot more information in the world than what you find on Pinterest or through AI.
For example, ACRNM set the trends for techwear, and the idea of academia began in Vienna in the 1900s.
Great games do this, they use authentic references. For example:
Deus Ex mixes the Renaissance with sci-fi.
Fallout is about the Cold War with a nuclear twist.
Witcher uses medieval stories from Slavic folklore.
Wolfenstein takes place in WWII but adds some extra history.
Curating Our Reference for Project Impact
Artists curate their ideas and inspirations for a project, like setting up a gallery in a museum. What artists choose to look at, like art, books and stories, influences our final work. This is the difference between copying and inspiration.
If an artist only looks at anime, their art will be different from someone who admires famous painters like Rembrandt and Da Vinci.
The artist's job is to curate looks, colors, shapes, and textures that communicate a message through images, without using any words at all.
The goal here was to create immersion, so I curated what I love to draw the most: the celebration of everyday life through images. This will support our game, because it'll add immersion for our players.
Students often ask me how to balance their own style with what a client wants, and I say it’s important to find a middle ground, to read the brief, until you find a place where you can be you and use that make the client project even better.
I used Photoshop to paint on top of the AI images the studio had, and fix the mistakes AI had made, and later, tested some AI features the new version Photoshop has, like the Generative Fill. I assisted the team with human-led design and story ideas to help them improve their project pitch.
Creating Project Immersion
and Human Connection
Instead of just showing nice buildings, I suggested we show people using spaces.
What I found was that this was a main ingredient in making the images approved faster. People connect to people, and we want to experience stories; that's why we like art in the first place. It doesn't need to be something big, but as long as it has intention.
This is how I curated my own solutions, one that I like and one that’d help the project stand out. I’m a fan of Zen and Buddhist philosophies, and I’ve always thought that snapshots of everyday life can be virtuous, when they don't try to be. It doesn’t need to look refined or pixel-perfect. Authenticity is attractive.
Does AI hold no Place in Your Production? That's a stretch.
I tried to find productive uses for AI, this was one of the tasks the studio assigned to mr, and found a few options where I could insert the references I wanted right into the environment and pose them.
It'd take me at least 1-3 hours to do the same by hand, and in this process, it took around 5 minutes, which is certainly faster than what I’d do, but remember, inserting already approved creative work in an already approved environment is a chore.
This is intern work.
Creative and Brainstorming is important work.
In the next images, you’ll see the reference AI inserted into the frame, but also the manual labour required to make it work, and how AI failed to understand the purpose of the task, and a human artist still had to manually review and interpret its work.
I was responsible for making the environments more appealing by curating and inserting those kinds of stories in the frame, which would suggest an RPG with depth of immersion and experience.
Image generated from the reference I provided to an AI.
AI did excel at placing the subject in the composition with good color and lighting fundamentals, but as you can see, there are lots of things not going right with the image, which I fixed by painting them over.
I helped fix the picture by painting it and making it better for our vision.
I decided that two people were enough to show the post-apocalyptic feel. I removed the plants, turned their clothes black, and used rags to show they were of a lower social class.
There was too much red and too many colors, so I kept the red to frame the main interaction.
I learned to use color to frame the image, after studying the "Pollice Verso" painting by Jean-Léon Gérôme, who also used red to help guide the viewer's eye.
The challenge was to find the balance between speed and not giving up our story; you can’t just deliver to a deadline, but achieve something meaningful.
The uncomfortable truth for AI enthusiasts is that AI didn't give me a great-looking result. They’d have to spend a long time fiddling with other AI tools to sort these artifacts, such as ComfyUI. However, I know how to draw and paint from scratch, so I didn't need to mess with complicated AI tools to deliver the work and get it approved on the first submission. I’d paint over it.
The process of not using AI for me would go like this:
One to three hours creating clothes in Marvelous Designer.
Posing the Characters in Daz Studio in 15 minutes.
Render the scene in KeyShot in 10 minutes.
About 30 minutes to an hour to overpaint the 3D.
When using Krea AI and Stable Diffusion, I inserted the photo reference in the scene, and it took 5 minutes. The base wasn’t perfect.
The main problem with AI is that everyone has it.
There's a lot of talk on whether AI will replace jobs, but I think those who are over-reliant on a single tool are those most vulnerable to being replaced. In my opinion, only using AI for a job is not enough to do good work.
In the example above I tried a few ways to “AI generative fill” my sketch.
AI didn’t fill in any knowledge gaps for my team or for me. It needed human reviews. I’ll show you a few. It has so far proven good at replicating existing content, but when you introduce a new challenge, it’ll break down.
Let's study this image before accepting that AI did fix it.
Looks impressive! Right?
Well..
As you can see, my 100% human-made sketch of the cloth showed a more efficient flow than an AI and a Human combined. The problem was that when AI tried to "fix" it ended up trying to "replace" what I had done with a detailed object, rather than improve upon what I made, so the ability to read results, whether they are from humans or AI, will always be an important skill, having the fundamental knowledge will help you interpreting these better.
I get why some people use it. The design on the red cloth would take longer to make by hand, but we could have added a special pattern. I’d rather draw them myself, but when working with a team, we need to be realistic. We should work together instead of pushing our ideas on everyone else.
Dali said, “Fear not perfection, you’ll never reach it.” It’s fine if AI comes in and adds imperfect but polished enough elements to call it good enough.
The previous "fix" suggested by AI is more polished, but even if glass can shine, it doesn't mean it is a diamond.
A pretty image with a dull idea will not be as scalable as an ugly image of a great idea. Brainstorming and researching are about using the ugly to find the great; it is not about finding a polished solution from the get-go.
I decided to find a balance between using AI and human skills. It was important to meet our deadline this week since we had a lot to do. While artistic vision matters, it should help the project, not hinder our goals. We're working together as a team.
I painted over this image using Photoshop.
Before that, I often create simple guidelines to pitch a revision and check with my clients to ensure we're on track. This helps us flag any challenges early and come up with new solutions before we face any problems. It also allows us to see if we’re meeting our goals and decide whether to keep going or change direction. In art for games or films, design is important. A polished picture doesn’t matter if we can’t use it.
The goal was to make our walled city more oppressive.
Overall, AI gave my client a good but superficial base.
The lighting was there, the anatomy looked right at first, but as you'll see, it breaks down under pressure.
We need analysis first, before execution, like a thief doesn't just break into the house, they plan. This is an analogy I learned from Richard Schmidt's Alla Prima book, the best book I have read on how to paint.
Analysis and execution are two parts of the design process, and it's a mistake to try to do both with AI.
Original by an AI artist with feedback from Miguel Nogueira.
Design theory fixed and painted over by Art Lead: Miguel Nogueira
Software Used: Photoshop, Krea AI, Stabble Diffusion.
Role: Senior Artist and Art team leader.
Software Used for Reference Gathering and Research: Miro, PureRef.
Time Spent: Design research could take from 30 minutes to 4 hours.
Paintovers would take around half the research’s time, depending on the image complexity.
Client Project Results:
The Images designed to improve the pitch deck's chance of approval.
All images delivered were approved for the pitch deck on the first try.
Images were used in the art guidelines for our project, and for future reference, as well as to instruct other team members.
Do you want to scale your vision with art?
Let me know in the button below, and help you get unstuck.