Skip to content

CREATIVEAILAB

Case study: Uncertain Tides

Breakdown

AI tools are revolutionizing not only storyboarding and pre-production but the full creation pipeline for immersive installations. Uncertain Tides / Onzekere (ge)tijden, developed by studio.POC in collaboration with the Flanders Marine Institute (VLIZ), is a striking example of how AI-driven workflows can be harnessed for narrative, emotional, and ecological storytelling.

This large-scale audiovisual experience was designed for a two-room bunker installation. In the second space, visitors are immersed in a surreal, destabilizing visual journey that explores the effects of ocean warming on marine ecosystems – especially the smallest among them: plankton.

At the heart of Uncertain Tides lies a tension between beauty and warning. The installation invites viewers into a gradually intensifying visual journey where soft, organic tones give way to vibrant, almost hypnotic colors. This shift is not merely aesthetic – it reflects a real ecological shift observed in marine research. As ocean temperatures rise, species such as bioluminescent algae (like Noctiluca scintillans, or “sea sparkle”) and jellyfish proliferate, introducing bright, flashy hues into the ecosystem. While visually stunning, their presence can signal imbalance and disruption. By using generative AI to mirror this evolution – allowing the visuals to grow more vivid and unstable over time – the artwork transforms scientific data into a sensorial narrative. Here, technology becomes more than a tool; it becomes a medium for ecological storytelling, translating unseen marine dynamics into a powerful emotional experience.

Let’s take a closer look at how this experience came to life using generative AI and a carefully crafted visual pipeline.

The foundation for the project was the Flanders Marine Institute plankton database, consisting of over 8 million images of microscopic organisms. These images formed the scientific and aesthetic backbone of the installation.

While traditional methods would struggle to translate this raw data into emotionally resonant imagery, generative AI provided a powerful alternative.

From: Flanders Marine Institute database © Flanders Marine Institute (VLIZ) 2025

To generate original visuals that were both inspired by and diverging from real-world data, the team trained a custom Stable Diffusion Lora-model using curated subsets of the VLIZ dataset.

This allowed studio.POC to maintain a consistent visual language throughout the installation, while also introducing interpretive, speculative imagery that stretches beyond the limitations of scientific visualization.


The visuals generated by the custom Stable Diffusion model served as the foundation for further stylistic exploration in Midjourney. Each Midjourney prompt began with a set of four reference images, anchoring the aesthetic direction.

With Midjourney’s intuitive prompt interface and strong stylistic cohesion, it became possible to:

  • Introduce surreal textures and organic forms
  • Evoke dark marine atmospheres
  • Visualize ecological instability through color shifts and composition

Visuals were generated using a custom ultra-wide aspect ratio (7.11:1) to fit the projection setup in the installation space.

Make sure to check out our Midjourney Guide

After generation, images were formatted and refined in Adobe Photoshop:

  • Cropped or filled (generative fill) to match the ultra-wide projection surface
  • Enhanced for optimal light projection (contrast, color balance)
  • Saved as frames to animate in Runway

Each generated image was animated using Runway Gen-2, an AI video tool that allows stills to become cinematic sequences via text or motion prompts:

  • Prompts emphasized destabilization and temperature shifts (e.g. “The ocean is boiling”)
  • Subtle camera movements (zoom, drift, pan) enhanced immersion
  • Clips were exported in short sequences, given the limitations of the technology (4 seconds)

An unintentional discovery emerged: when clips were extended repeatedly in Runway, the visuals began to “break apart”. Instead of correcting this, the team embraced it as a poetic metaphor for ecological collapse.

This breakdown became a narrative technique: as water temperatures rise, the microscopic marine world begins to unravel visually—a striking symbol of ecological fragility.

Before the editing phase, all animated clips were processed in Topaz Video AI to enhance their motion quality and visual fidelity:

  • Clips were slowed down and upscaled to projection-ready resolution
  • Additional frames were generated to ensure smooth, fluid animation
  • Noise was reduced and visual clarity was preserved even at large scale

This ensured that the delicate, drifting textures remained sharp and immersive when projected across the installation’s four surrounding walls.

With hundreds of clips in hand, the final sequence was edited in Adobe Premiere Pro:

  • Sequenced by narrative arc: a slow world-building start, evolving from balance to breakdown
  • Aligned with an original score by Charo Calvo
  • Crossfades and camera logic created seamless spatial flow

Here’s the final result:

Takeaways for your project

Uncertain Tides demonstrates how generative AI can be a powerful tool for immersive storytelling, artistic experimentation, and scientific interpretation. Here are some key takeaways for creators working with AI:

  • Work from real-world datasets. Starting with existing image collections – like the VLIZ plankton database – anchors your generative process in meaningful content and offers strong visual cohesion.
  • Use AI tools in dialogue. Combine different platforms for different creative needs: use Stable Diffusion for control and customization, Midjourney for stylistic variety, and Runway Gen-2 for cinematic motion.
  • Leverage prompting as creative direction. The way you prompt – especially when using visual references – can shape the output as precisely as a storyboard or mood board would.
  • Embrace unpredictability. AI models often introduce unexpected artifacts or shifts. Rather than seeing them as flaws, consider them as opportunities to inject symbolism or narrative depth into your work.
  • Think in iterative loops. AI workflows thrive on testing, remixing, and refining. Build your project in short cycles, allowing each stage to inform the next.
  • Post-process purposefully. Upscaling, slowing down, or reanimating AI output can dramatically increase its cinematic and emotional impact – especially for installation-based or large-format work.

Generative AI is not just a tool for creating visuals – it’s a collaborator that can push your imagination beyond traditional boundaries.