"Jutters"

AI summary

Overview

Research area: Human-Computer Interaction, specifically AI-generated media, curation, and interactive art installation design.

Technical level: Beginner-Friendly. The paper is a descriptive account of an art installation; it contains no formal system architecture, no algorithms developed by the authors, and no experimental evaluation.

Scope: A one-page-style project description of "Jutters," a physical beach-like installation that relocates AI-generated images and AI-transformed videos into a sandy space where visitors act as Dutch coastal foragers deciding what to keep and what to discard.

What This Paper Is About

Most digital content we encounter is served to us rather than sought out, and much of it is now AI-generated, which makes it hard to tell what is original, real, interesting, or purposeless. The authors argue that AI-generated content resembles objects washed ashore: it lacks clear authorship, history, ownership, and intended use. Their goal is to reframe this problem through the older Dutch tradition of the jutter — coastal foragers who gathered and repurposed what the sea left behind — by building an installation where visitors physically sort through AI-made objects and decide what is worth keeping.

Key Contributions

1. A conceptual reframing: The paper proposes the jutter tradition as a lens for thinking about how people should engage with AI-generated content, emphasizing curation, judgment, and repurposing rather than passive consumption.
2. A physical installation design: An approximately 2 × 3 meter sand-covered space containing real tidal debris from Noordwijk beach in the Netherlands, 7 framed prints of AI-generated images, and 2 screens showing AI-transformed videos.
3. A documented generation pipeline: A description of how the AI imagery and video were produced, including over 100 photographs of objects found on the shore of Hoek van Holland beach, image generation with GPT-4 Omni, video transformation with WarpFusion, a Stable Diffusion v1.4 baseline, and TopazLabs "Frame Interpolation" post-processing.
4. An accessible public artifact: A narrated, walkable experience in which participants are guided by audio narration and wave sounds, plus a publicly available video preview at https://www.youtube.com/watch?v=L6319Ii7MT8.

Main Findings

• The conceptual parallel: The authors position AI-generated content as analogous to shipwreck debris — cargo, wood from sunken ships, and trinkets fallen overboard — in that it arrives detached from authorship, history, ownership, and intended use. They cite the jutter tradition as documented in reference [17].
• The volume of AI content is framed as a problem of discernment: The paper states that a large part of the content we consume digitally is now AI-generated (citing [22]) and that we increasingly struggle to distinguish what is original, real, interesting, or purposeless (citing [12, 13]).
• Installation composition: Roughly 2 × 3 meters of sand, with lighter-coloured sand marking a path and darker-coloured sand in the remaining areas; some parts dry, others humid. Real shells, small driftwood, and tidal debris were collected from Noordwijk beach, a site the paper describes as once frequented by jutters.
• Generated media counts: 7 framed prints of AI-generated images and 2 screens showing AI-transformed videos, alongside the real debris.
• Image selection criterion: From images generated with GPT-4 Omni based on pairs or triplets of photographs, the authors selected images that "appeared familiar but uncanny," intending to evoke ambiguity and invite interpretation, so participants would make fresh judgments unanchored by prior associations.
• Video generation outcome: Final videos were produced with WarpFusion, described as an AI video-to-video transformation model that distorts form and motion in a surreal but fluid style. Stable Diffusion v1.4 served as a baseline because the authors found earlier diffusion models generate more surreal imagery.
• No formal evaluation is reported: The paper presents no user study, no participant counts, no measured outcomes, and no comparative benchmark results.

Methodology in Plain English

The authors built a small indoor beach and then populated it with two kinds of material: genuine debris and AI-made stand-ins.

For the real material, they collected shells, small driftwood, and other tidal debris from Noordwijk beach in the Netherlands. For the AI images, they first photographed over 100 objects found on the shore of Hoek van Holland beach, then used GPT-4 Omni to generate new images based on pairs or triplets of those photographs, and picked the results that felt familiar yet uncanny. For the videos, they recorded footage at Zandvoort beach, tried several video-to-video AI models, and settled on WarpFusion to create videos of objects morphing while being rocked by waves, with Stable Diffusion v1.4 as a baseline model. They used the TopazLabs "Frame Interpolation" tool to raise the frame rate and made minor colouring adjustments manually in video editing software.

Visitors walk along the lighter sand path while an audio narration with wave sounds introduces the jutter tradition and asks them to reflect on lost histories, altered meanings, and possible future roles for these digitally reshaped fragments. The installation video is in Dutch with English subtitles, while the narration for the installation itself was translated into English.

Why This Matters

The work matters less as a technical advance than as a design provocation: it argues that curatorial judgment is becoming a core skill for living with AI-generated media, and it offers an embodied, physical way to practice that judgment rather than a screen-based one.

Real-world applications:

• Museum and gallery programming: A ready model for exhibits that ask visitors to physically handle, sort, and judge AI-generated material.
• Media literacy education: A classroom or workshop format for teaching audiences to interrogate the provenance and purpose of synthetic media.
• AI ethics and design workshops: A hands-on exercise for designers and developers thinking about provenance, authorship, and user trust in generated content.
• Cultural heritage engagement: The project's use of public archives — the Netherlands Institute for Sound and Vision, the Regional Archives of Alkmaar, and the Juttersmu-ZEE-um museum — shows a way to connect AI art with historical collections.

Industry relevance: For platforms that serve AI-generated content, the paper's framing suggests that the design problem is not only detection but also curation — helping users decide what deserves attention. The generation pipeline described (GPT-4 Omni for images, WarpFusion for video, Stable Diffusion v1.4 as baseline, TopazLabs frame interpolation for post-processing) is also a practical recipe for artists and creative technologists producing deliberately uncanny synthetic imagery.

Future Directions

• Evaluating the visitor experience: The paper reports no data on how participants actually behaved — what they kept, what they discarded, or what they thought. A study measuring these decisions would test the authors' claim that the installation encourages more discerning engagement.
• Testing the curator role empirically: The authors propose that a "crucial new type of curatorial role" is emerging in the age of AI-generated content; whether people can or do adopt this role remains an open question.
• Scaling and varying the installation: Only 7 prints and 2 screens were used in a roughly 2 × 3 meter space. Larger or repeated installations could reveal whether the effect holds with more material or different configurations.
• Extending beyond the Dutch context: The reframing depends on a specifically Dutch coastal tradition; whether the jutter metaphor translates to other cultures and other digital-media ecosystems is unexamined.

Target Audience

This paper benefits most from HCI researchers and interaction designers interested in AI and culture, artists and creative technologists working with generative models, museum and exhibition curators, and media literacy educators. General readers curious about how to think critically about AI-generated imagery will also find it accessible, since it contains no technical prerequisites. Readers looking for quantitative results, benchmarks, or formal user evaluation will not find them here — those are not reported.

Authors’ abstract

This project explores how we engage with AI-generated content through the lens of the jutter: Dutch coastal foragers who comb the shoreline after storms, gathering and repurposing what the sea leaves behind. Reflecting how our lives are increasingly shaped by AI-generated media, we create a beach-like installation that blends real shoreline debris with AI-transformed images and videos. Visitors are invited to explore this space as contemporary jutters, deciding what to keep and what to discard. In doing so, the project reimagines AI-imagery as material for reflection, encouraging a more discerning engagement with the content that drifts through our feeds. A video preview of the installation can be found at https://www.youtube.com/watch?v=L6319Ii7MT8.