AETHER- Visual Poetry Engine

Inspiration

The project is inspired by visual poetry, slow cinema, and systems art—particularly practices that emphasize reduction, repetition, and temporal tension over narrative clarity. It draws from the idea that meaning often emerges not through motion itself, but through the viewer’s perception of difference across still states - especially the experience of ADHD, where thoughts and feelings rarely progress linearly.

Rather than forcing AI to generate cinematic motion, the work asks a different question: What if emotional transformation looks like variation, not movement?

The imagery intentionally sits between stillness and motion, creating a quiet tension that invites the viewer to project their own inner rhythm onto the work.

What it does.

Aether is an experimental visual poetry engine that investigates how generative AI can support continuity and meaning rather than isolated outputs. Instead of treating AI as a tool for producing individual images or clips, the project frames generation as a temporal process—where symbolic motifs, tone, and visual rhythm persist across states. The system emphasizes constraint, reduction, and sequencing over volume, asking how much meaning can be preserved when variation is intentionally limited.

Aether uses Gemini’s generative image capabilities as part of a constrained creative pipeline focused on continuity rather than novelty. Gemini is prompted with poetic, non-literal descriptors designed to generate painterly visual states instead of literal scenes.

Rather than maximizing variation, the system intentionally limits prompt drift so that each generated image functions as a transformation of the previous state, preserving symbolic and tonal coherence across a sequence. This approach treats Gemini not as a one-off image generator, but as a collaborator in a temporal process—where meaning emerges through accumulation, repetition, and subtle mutation.

Due to practical generation limits, the project works with a small, curated set of outputs, emphasizing selection and sequencing as core creative acts. This highlights an important design insight: effective use of generative AI often depends less on scale and more on constraint, editorial judgment, and intentional framing.

Aether demonstrates how Gemini can support artistic workflows that prioritize authorship, pacing, and conceptual integrity over raw output volume.

How we built it

The system was built as a lightweight visual interface that presents curated generative sequences rather than live output. Generated images are organized into states and played back through timed transitions to emphasize temporal continuity.

A key design feature is a logic toggle that reveals how prompts evolve across states. This allows users to observe the underlying prompt transformations that guide each generative step, making the system’s internal logic visible rather than opaque.

By exposing prompt changes instead of hiding them, Aether treats prompting as a compositional layer—inviting reflection on how language, constraint, and structure shape generative outcomes. Generated images are organized into states and played back through timed transitions to emphasize temporal continuity. The interface remains deliberately minimal to keep focus on sequence, rhythm, and visual residue rather than UI spectacle.

Challenges we ran into

A primary challenge was maintaining visual and symbolic continuity across generative outputs, as AI models naturally favor reinterpretation over memory. Another challenge involved working within strict generation limits, which required treating selection, sequencing, and framing as first-class design decisions rather than post-processing steps. Making the system’s internal logic legible without overwhelming the viewer required selectively exposing prompt evolution through a toggle rather than default visibility. These constraints ultimately shaped the project’s methodology and aesthetic, reinforcing the idea that limitation can function as a creative signal rather than a deficiency.

Accomplishments that we're proud of

Firstly resisting the expectation that AI-generated visuals must be hyper-cinematic or technically impressive.

Early iterations produced images that felt uncanny or overly dramatic, which conflicted with the subtle emotional tone of the project. The challenge became learning how to limit AI—using it not as a spectacle engine, but as a mindful collaborator.

Another challenge was consistency across generations, as different tools interpret prompts and frames differently. Instead of hiding this inconsistency, the project integrates it as part of the emotional language. Proud of framing generative AI as a system of continuity and authorship rather than output volume, and of exposing prompt logic as part of the creative structure instead of hiding it.

What we learned

The project reinforced that constraint, curation, and transparency often matter more than scale when working with generative systems, especially in time-based creative contexts.

What's next for AETHER- Visual Poetry Engine

Future development would focus on extending continuity across longer temporal spans through memory mechanisms, structured prompt inheritance, and adaptive sequencing logic.

Additional directions include deeper multimodal integration and expanded transparency tools that allow users to actively shape or annotate prompt transitions as part of the creative process.

Built With

• ai
• gemini
• studio