THE THING WE BUILT:

Mocktopus

My first design internship involved a lot of photoshopping.

Photoshopping a logo onto a mug, a billboard, a tote bag.

When the team was working on an exhibit, it got even harder: putting an object in a glass case. Making sure the reflections were perfect. Mocking up 20 different labels in the space.

There was a senior designer on my team that was a pro at this. I remember watching in awe as he stacked layers, feathered edges, cloned the background to look just right.

I spent a lot of time trying to get really good at this. But it was tedious.

The last few weeks it felt like the focus has been on AI companies creating new models to make even sloppier slop generative images and videos (and now even to lock you into their own social platforms.) That’s for another newsletter, but perhaps lost in that conversation: Gemini’s 2.5 Flash Image model is…. actually doing a good job with images? Instead of disregarding the image you upload just to regenerate an uncanny valley version of it, it seems to stay pretty faithful to it. And can manipulate it, combine it with more than one image you upload. Sounds a lot like my intern days.

So we built a thing.

Mocktopus in action

How It Works

1. Choose one of three different modes:

   • Slide Mockup - combine two images together based on your instruction

   • Dynamic Variations - create variants on the image you upload

   • Style Transfer - transfer the style of an image to your original graphic

2. Upload relevant images and describe what you want

3. Get mockups that actually use YOUR graphic, not an AI fever dream or the uncanny version of what you were trying to design

4. Download the results without spending your afternoon masking and warping

It feels like operating Photoshop with 8 arms at once (or maybe I just really like the Mocktopus pun)

Notes from Building

Gemini 2.5 Flash Image is the first generative AI model I’ve used that actually feels like it respects the image you upload. When you say “use THIS graphic” and “put it on THIS background” you're getting 1:1 results.

Using generative models like DALLE or Stable Diffusion have been fun, maybe even useful for rough sketches, but trying to use them in real presentations or even production has felt like a stretch. Suddenly, that’s different. Building on top of 2.5 Flash Image, I can actually see this replacing much of what we do in Photoshop now. Fast, high quality, worth sharing quickly with clients.

It also makes visual continuity possible. We can take a mascot, generate variations, then generate variations of those variations without it turning into a horrifying game of visual telephone. Your spaceman stays a spaceman. Your brand colors stay your brand colors. Perhaps it doesn’t sound revolutionary. But apparently it is.

I’m sure we’ll still find ourselves photoshopping some of the finer details, but watching the most tedious parts of this work disappear is wild. And clearly just the start. While I lament that my photoshopping skills may not be quite as valuable as they used to be, I’ll be pondering what we can do with this next.

Built with ❤️ by Township

You’re receiving this because you signed up for Township’s “We Built a Thing” newsletter. We'll only send these when we, well, build things.