Check out Part 1 here:
We keep talking about “Shadow Doll.” And we’re asking people to watch Eva and Hailey review video games – what if we made a fake video game called Shadow Doll? Vermis did it and now everybody makes art of video games that don’t exist.
– Twilight
That’s a terrible idea.
– Shelby Moore
In retrospect, it was a terrible idea.
Not conceptually – we make AI art – in terms of what is possible, there are very few limitations that can’t be solved creatively.
But, you need the right tools for it – and you need to understand just how much work this actually is.
To say I was not prepared would be an understatement.
The box art came out great.
This was the prompt we used for the American Art:
A box cover for an unlicensed Nintendo Entertainment System game called Shadow Doll. The cover features a stylized painting of an athletic female with straight blue hair with bangs holding a laser katana and standing in front of a wall of biomechanical technology. Strong blue and green color tones
And I have to say that looks REALLY period accurate as to what box art for a Nintendo game might have looked liked in the 1980s and 1990s.
For the sake of comparison, let’s look at how Perchance [] handles that same prompt:
While a lot of that art is more like what you might have found inside of an instruction manual or possibly on the pages of Nintendo Power, it definitely looks the part. You can see that Perchance struggles a bit with the logos and text.
Naturally, the European box art would have a bit more edge to it:
A box cover for an unlicensed Nintendo Entertainment System game called Shadow Doll. The cover features a stylized painting of a nude athletic female with straight blue hair with bangs holding a laser katana and standing in front of a wall of biomechanical technology and monsters fused to the wall and reaching out towards her. Strong blue and green color tones
Part of why there are so many generations of that is because I really loved the style and the imagery that we were getting with that particular prompt.
And Perchance did a great job with the same prompt:
What I find really interesting about the Perchance generations is how you can see a Moebius and Richard Corben influence in them – which although it’s not stated in the prompt, it’s very appropriate for the European box art.
There was one thing that the image was missing though: a gameplay gimmick.
In this case, our gameplay gimmick was “biomechanical fusion” – the ability to absorb weapons and tech from fallen enemies. So we needed a prompt to reflect that:
A box cover for an unlicensed Nintendo Entertainment System game called Shadow Doll. The cover features a stylized painting of a nude athletic female with straight blue hair with bangs holding a laser katana and standing in front of a wall of biomechanical technology and monsters fused to the wall and reaching out towards her and fusing her body with the same biomechanical technology. The box notes that it’s an “ultra cartridge” featuring the “SD graphics chip” a “laser katana” and “biomechanical fusion.” Strong blue and green color tones
A box cover for an unlicensed Nintendo Entertainment System game called Shadow Doll. The cover features a stylized painting of a nude athletic female with straight blue hair with bangs holding a laser katana and standing in front of a wall of biomechanical technology and monsters fused to the wall and reaching out towards her and fusing her body with the same biomechanical technology. Her back is to the viewer. The box notes that it’s an “ultra cartridge” featuring the “SD graphics chip” a “laser katana” and “biomechanical fusion.” Strong blue and green color tones
THAT was the Shadow Doll look – and the last image was selected to be the canon box art for the game.
Perchance did not seem to understand the idea of biomechanical fusion:
I’m not sure what happened there, but the quality definitely took a hit.
Now we needed the Japanese box art:
A box cover for an unlicensed Nintendo Entertainment System game called Shadow Doll. The cover features a stylized manga painting in the style of Masamune Shirow of a nude athletic female with straight blue hair with bangs holding a laser katana and standing in front of a wall of biomechanical technology and monsters fused to the wall and reaching out towards her and fusing her body with the same biomechanical technology. The box notes that it’s an “ultra cartridge” featuring the “SD graphics chip” a “laser katana” and “biomechanical fusion.” All text is in Japanese. Strong blue and green color tones
The Shin Megami Tensei vibes with this are pretty incredible.
Let’s see what Perchance made with the same prmopt:
While it’s not the same quality as Grok, I do like the aesthetic choices and while some are rought, sokme are very well done.
Since we had our box art, now it was time to make an instruction manual. It was a fairly straighforward process, using the box art as a base image, these were the prompts:
Using the base image as a reference, create page 1 of an instruction manual maintaining the same art style. Use this prompt: Page 1: Welcome to Mega City [at the top of the page is an illustration of a cyberpunk city skyline at sunset with dominant orange and pink color tones] The text reads: A doomed megalopolis rising up out of the remains of what was once the west coast of the United States; the last major city still left standing in 19XX; this is Mega City. The neon lights fast fade in a city where life is cheap. Nothing is free and everything has a price. You either rise through the corporate ranks or you make your name in the streets. [at the bottom of the page is an illustration of a street gang with cybernetic implants wearing clown masks. They are standing on a street corner]
Using the base image as a reference, create page 1 of an instruction manual maintaining the same art style. Use this prompt: Page 2: The Story [at the top of the page is an illustration of the female in the reference. She is wearing a skimpy bikini top and a red flight jacket with padded leather pants. She is holding a small cyberdeck-style computer with punk rock and anti-corporate stickers on it. She is sitting on a futuristic motorcycle in a casual pose] You are a netrunner named Nova. The netrunners are the shadow warriors in corporate warfare. Hacking systems. Stealing secrets. Hiding in the shadows. Always looking over your shoulder, but you can’t quit because that next big data mine could be right around the corner. You and your boyfriend Dusk came up through the streets. Doing small time hacks for street gangs leeching credits from 24 hour markets. Then, as your reputation grew, you had corporate clients who paid you a lot of money to steal secrets from their competitors. You were good at your job. But you made the wrong decision with the last job you took You hacked the OROMO corporation’s Motherbrain mainframe and stole corporate secrets worth millions of credits. Not just design plans and code; you found blackmail videos; hit lists; hidden surveillance feeds and phone wiretaps. That data and access codes were worth millions of credits – you and Dusk were going to be set for life. And then the OROMO corporation found you. [there is an illustration at the bottom of the page of men in black suits standing at the entrance to an apartment building in the slums. They are flanked by private security wearing exo-skeletons which have the branding of OROMO on them]
Using the base image as a reference, create page 3 of an instruction manual maintaining the same art style. Use this prompt: Page 3: [at the top of the page is an illustration of the girl being dragged out of her bed naked by OROMO security wearing exo-skeletons which have the branding of OROMO on them. In the background men in black suits watch.] OROMO came in the night and dragged you to one of their underground laboratories. They stripped you of your cybernetic implants and then shoved new tech inside of you with new programming. After hours of torturous surgery without anesthetic, you were no longer Nova – you became a Shadow Doll – a corporate slave-assassin wired to kill. [there are 2 illustrations at the bottom of the page. Illustration 1 shows the girl in the reference being fused with new cybernetic implants. Illustration 2 shows the girl in the reference standing on a rooftop and holding a laser katana – in the background is the skyline of a cyberpunk city at night]
Using the base image as a reference, create page 4 of an instruction manual maintaining the same art style. Use this prompt: Page 4: [there is an illustration at the top of the page that shows the girl in the reference using a laster katana to kill a man in a black suit in front of a corporate office building in a cyberpunk downtown area] You killed a lot of people You were good at it. But there was still a part of you that remembered who you used to be. OROMO couldn’t control you. Not really. And one day you broke free. A power blackout knocked out security services for half the city and you managed to sever your connection with the OROMO Motherbrain. Now, armed with the ability to fuse your body with enemy tech, it’s time for revenge. You’re going to destroy the OROMO Motherbrain. And you’re going to kill everyone that gets in your way. [at the bottom of the page is an illustration of the girl in the reference with additional cybernetic armor, including a jetpack that looks like wings. She is holding her laser katana and standing in front of a massive corporate building with the logo of the OROMO corporation on it.]
And that was great, but now we had to actually make a game…
And it’s not that Grok can’t do pixel art – it absolutely can.
The problem is that 1980s sprite art displkayed on a CRT screen has a very specific aesthetic, and while I can immediately recognize it, it’s difficult to explain to a machine that has never actually played an NES game.
I tried to phrase it in a way that made sense to me:
Create a mock up of the first stage of a video game featuring a nude athletes female with straight blue hair fighting mutants and cyborgs using a laser katana. Enhanced 8-bit style pixel art. The first stage is called Mega City. The game is a side scrolling action platformer similar to Metroid and Castlevania
This was the initial output:
So it was pretty clear that Grok just doesn’t understand the aesthetic – which makes sense – not many people are trying to use AI to make things that look like old video games.
In defense of Perchance, everybody’s favorite free image generator, it actually did a pretty good job with the same prompt:
None of that is “right,” but I do think Perchance has a better grasp of “8-bit” than Grok.
We tried an optimized prompt:
Generate a mockup screenshot of a classic 1980s NES 8-bit side-scrolling action platformer exactly like early NES games such as Castlevania, Metroid, or Contra. Strict authentic NES hardware limitations: 256×240 resolution feel, extremely limited NES color palette with only a few colors per sprite and background layer, chunky blocky low-resolution pixel art with large visible pixels, heavy dithering patterns instead of gradients, no smooth lines, no anti-aliasing, no modern shading or effects, pure crude retro 1985-1989 aesthetic. The first stage is called Mega City: dystopian cyberpunk city background with blocky skyscrapers, industrial pipes, platforms, neon signs, and streets rendered in simple tile-based 8-bit graphics. In the center of the screen, a nude athletic female protagonist with straight blue hair is dynamically fighting several mutants and cyborgs using a glowing laser katana in an action combat pose. At the top of the screen, classic NES-style HUD elements: life bar, energy bar, experience point counter, and the text ‘MEGA CITY’ in blocky 8-bit font prominently displayed. Overall visual style must be unmistakably authentic 8-bit NES, low-detail and pixelated. Rendered with a strong authentic CRT filter: visible horizontal scanlines, phosphor glow, slight screen curvature, RGB phosphor dot pattern, and retro TV artifacts as if the image was photographed directly from a 1980s CRT television monitor.
And this was what Grok gave us:
Which isn’t bad. I think that graphically, that’s about the look we were going for – that’s not quite NES level, but since the lore for Shadow Doll was that it’s an “enhanced” cart with a graphics chip in it, sure, that could work.
This is what Perchance made with the same prompt:
And I have to say that I am pretty impressed with what Perchance made. It applies a CRT filter differently, and based on the colors this is maybe more akin to either a Sega Master System game or some kind of PC game – but it looks like what it’s supposed to look like and the prompt adherence is pretty good.
You might have noticed there’s a lack of consistency in the game screens. And maybe if we just had still images we could muddle through it.
But there was also the issue of animating all of this.
Because Grok had a tendency to take an image that looked like this:
And make a video that looked like this:
And NOTHING about that works.
So once it became apparent that we were not going to create anything that was convincingly 8-bit, we shifted the focus to creating interesting images and situations. The in-world explanation was simple: Shadow Doll had an advanced graphics chip inside of the cartridge that generated high end graphics but also had a tendency to make the console catch fire from overheating – the Sudden Death chip.
Here’s everything that was generated while trying to piece together Shadow Doll footage:
And this doesn’t give an indication of the true scale – because there were literally thousands of images generated and scrapped as part of the process. So a process that should have taken a few hours turned into a few days.
Ultimately, I’m glad we did it this way – I think we had to be able to justify the gameplay footage as not just filler for Eva’s segments.
But there’s also a reason why Hailey’s segment relied on gameplay footage from a real video game.
In Part 3 we’ll talk about the creation of the Shadow Doll commercial.
In the meantime, check out some of our recent posts:
Leave a Reply