AI-Generated Deep Sea Landscapes: Beginner Questions From Someone Who Sails Over Them

Last month, I was standing the night watch in the waters south of Crete. On the sea above depths exceeding 2,000 metres, if you turn off the deck lights, you can sometimes see faint bioluminescence below the surface. Watching that glow, I always think the same thing: what does it actually look like down there? It's a place a camera could never capture. That's why I started using AI image generation tools.

I've had quite a few similar questions pile up through blog DMs and Instagram comments. Here's my attempt to address them.

"Does it matter which tool I use?"

Honestly, they all look similar at first. I started with Stable Diffusion and now use Midjourney alongside it. Both can create deep-sea landscapes, but they have different strengths. Stable Diffusion can run locally, which is an advantage in environments with unreliable internet — like on a ship. I've had more than a few timeouts trying to send Midjourney prompts over satellite Wi-Fi. On the other hand, Midjourney produces great textures without any separate installation.

What matters more than the tool is how specifically you can picture the scene you want to depict. The mesopelagic zone at 200 metres and the abyssal floor at 4,000 metres are completely different, starting with the amount of light. The tool can't tell them apart. You have to tell it.

"How should I write prompts?"

This is the question I get the most. I started out typing things like "deep sea landscape, beautiful, realistic" too. The results were pretty but looked familiar — like aquarium wallpaper.

Things changed when I started putting what I'd actually seen and felt on the water into my prompts. For example, here's the kind of combination I use often:

• Specifying the light source: "bioluminescent organisms as the only light source, no sunlight penetration"
• Describing texture: "volcanic basalt formations covered in pale bacterial mats"
• Implying scale: "a single remotely operated vehicle spotlight revealing the scale of a hydrothermal vent chimney"

When creating landscapes near hydrothermal vents, if you just write "hydrothermal vent," most AIs will mix in images of surface volcanoes. The trick is to explicitly exclude terrestrial elements, like "black smoker chimney on the abyssal plain, mineral-rich dark plume rising into cold water, no sky visible."

"How does it compare to actual deep-sea photography?"

I want to answer this one carefully. Images made by AI are not scientific records. I confused that myself at first. If you look at real deep-sea footage released by institutions like NOAA or MBARI, you can quickly see how much AI images distort things to make them "pretty." The real deep sea is far more monotonous, washed of colour, and precisely because of that, more frightening.

But that doesn't make it worthless. I see AI deep-sea images as a kind of impressionist landscape painting. Just as Monet didn't paint the Rouen Cathedral exactly as it was, the deep sea I create with AI is an impression of the world below that I imagined from up on deck. Where you place your own work on the spectrum between accuracy and expression is up to you.

Equipment and Costs

"Do I need an expensive GPU?" comes up a lot too. The personal laptop I use on the ship is a pretty old ThinkPad, and frankly it struggles to run Stable Diffusion. So while we're in port I use cloud-based services at cafés, and during voyages I just jot down text prompts and generate everything in a batch later. A basic Midjourney subscription is roughly around 10 euros a month, so it's not a burdensome expense.

You can start without a GPU. Cloud services or subscription-based tools are enough. If you plan to work on a ship, though, you need to seriously consider an offline setup. I keep a notebook where I write prompts by hand in case a voyage runs long. It's funny how an analogue habit ends up improving the quality of digital work.

"What can you actually do with this?"

I use it for personal short film projects. I cross-edit above-water footage shot during voyages with AI-generated below-water images. Deliberately blurring the line between reality and imagination is the point. Some people find it unsettling; others like the ambiguity. I think both are valid reactions.

Beyond artistic uses, there are interesting practical applications too. When explaining the seafloor terrain of a particular area to the captain I work with, I once showed him AI images to convey a feeling that nautical charts alone couldn't communicate. It lacked scientific accuracy, but it was effective at building an intuitive understanding — a sense of "oh, so this is how deep it is here."

I float on this sea every day and have never once seen what's below. I could just accept that, but I'd rather at least try to imagine it.