Creative Technology · System Design · Brand Strategy · AI Assisted Prototyping
A release system prototype shaped by sound and steered by the artist.
I designed Selkie as a workflow, not an image generator. Artist inputs move through a structured prompt system to create three distinct directions, then into controls for palette, typography, imagery, composition, motion, and export.
Experience the live prototype ↗Independent artists are expected to release a complete visual world alongside every track, often without an art director, production budget, or the language to brief a collaborator. The alternatives are usually disconnected DIY assets, generic AI imagery, or agency support priced beyond an emerging artist's reach.
Selkie organizes the track, release context, references, exclusions, and artist intent into a creative brief. Prompt rules translate that brief into three deliberately different worlds, each with its own concept, color roles, typography, image language, texture, and motion behavior.
The system proposes the creative territory. The artist chooses, combines, and redirects where it goes.
The artist uploads a track and adds release details, references, intended feeling, and exclusions. These inputs become constraints for the system rather than decoration around a generic prompt.
Prompt and workflow rules separate the response into three art direction propositions rather than treating the first generated look as the answer.
Through the World Mixer, the artist can keep a world intact or combine its imagery, palette, typography, and energy with another direction.
Artist visibility, image treatment, composition, palette, typography, and restrained, charged, or fractured motion modes shape the final system.
I tested the prototype with different songs and creative inputs, then evaluated whether the worlds were genuinely distinct, the controls changed the direction in meaningful ways, and the outputs felt usable beyond the screen.
Palettes, fonts, and image language began repeating across songs, which made the system feel stylistic rather than responsive.
RefinementI tightened the rules separating each world and defined color, type, imagery, texture, and motion as distinct system variables.
A strong first image was not enough. References could repeat, questions felt abstract, and early motion studies were too uniform or slow.
RefinementI added clearer inputs, exclusions, the World Mixer, cover controls, and motion modes so the artist could redirect the result instead of restarting.
The concept needed to move from inspiration toward assets an artist could actually use or hand to a collaborator.
RefinementI expanded the workflow into a cover builder, a color and typography system, motion output, and a downloadable creative brief.
Walk through the complete v0.13 flow from track and artist input through three world generation, World Mixer controls, responsive cover styling, motion modes, and downloadable outputs.
Artist references, exclusions, world selection, mixing, palette, type, imagery, and motion behavior all remain steerable throughout the experience.
Each session develops into a connected release system rather than a single generated image. In the Glimmer example, the artist can refine and download the cover, export an eight second motion loop, use a color and typography system, and hand a generated brief to a collaborator.
The proposed model keeps Selkie accessible to artists who release occasionally while creating recurring value for active artists and small labels. Pricing is structured around the real compute cost of AI generation and motion rendering. It does not rely on unlimited generation disguised as free creative.
A focused release kit for one single or campaign, without an ongoing commitment.
Recurring generation, refinement, and export capacity for active artists and small labels.
A tiered model designed to cover API and rendering costs while expanding margin as the product scales.
In folklore, a selkie moves between two forms. The name became a metaphor for the product itself, an experience designed to translate sound into image while preserving the feeling that connects them.
I conceived Selkie and built the product logic, input flow, prompt architecture, three world framework, artist control system, visual identity, web interface, motion behavior, evaluation criteria, and deployed prototype.
The current prototype includes artist inputs, AI assisted world generation, steering, responsive cover updates, motion previews, and downloadable outputs. It is not yet a production grade audio analysis or commercial release platform.
I used an AI assisted development workflow to turn the concept into a working web experience. My contribution centers on product logic, creative direction, interaction decisions, prompt design, testing, and deployment. It does not represent production level software engineering.
The stack combined creative direction and interface design with AI assisted development. Each tool supported a specific part of the workflow rather than standing in for the idea.
The next product phase would add production grade audio feature analysis, final Spotify ready MP4 encoding, adaptable social templates, batch exports, and collaborative review. This would turn the current art direction engine into a complete release campaign workflow.