Min Ye-Eun

This video work begins with a painting (Untitled) by Ayako Rokkaku and combines image, music, and character-based narrative into a single unified audiovisual frame. Drawing from a selected work encountered in the artist’s solo exhibition at Total Museum, key visual elements—color, rhythm, density, and repeating forms—are translated into musical structures to generate a dreamlike background score. At the center of the screen, a character derived from the painting is reconstructed as a live-action figure. This character, configured as a being who speaks Korean, German, and Japanese, delivers a poetic monologue about the “planet” they come from and their purpose in this world. The multilingual narration is realized through AI-generated voice, forming a cohesive audiovisual scene in combination with generated images and video.

The work originates from the question: “To what extent can a single painting be translated into other media?” Rather than treating the selected Untitled painting as an object of passive viewing, it is redefined as a source that generates rules for music, speech, and movement. The image is first mapped onto a musical system to produce sound, after which the central figure is rendered as a live-action character and framed as an entity from another world. The use of three languages reflects the premise that a single world cannot be reduced to a singular linguistic system.

In the production process, only the initial direction and parameters were defined by the artist, while the iterative development of prompts was largely delegated to AI in order to approach the desired outcome. The decision to also co-author the description of the work with AI reflects a broader conceptual interest in authorship—who creates and writes—and in the chain of translation (image → score → voice → video → text) as an integral part of the artwork itself.

Work Information

• Title: Order Found in Chaos: Star Child 1
• Artist: Min Ye-eun
• Format: MP4
• Duration: 40 seconds
• Aspect Ratio: 16:9
• Resolution: 1904 × 1072
• Frame Rate: 30 fps
• Codec / Bitrate: 11657 kbps
• Audio: Stereo, 2-channel

Dataset (Provided by Total Museum) and Selection Scope

• Source: Dropbox (including exhibition texts, artwork information, and images)
• Primary Input: rokkaku_Japan-Korea_1 (single artwork image)
• Supplementary Reference: Full review of exhibition-related text files (for defining conceptual conditions)
• Additional Research: Web-based research on Rokkaku (reference only; core generation is based on the selected dataset image)

AI Models and Services Used

• Image Generation: Kling 3
• Video / Digital Character: Kling 3
• Voice Synthesis (TTS): ElevenLabs – Multilingual v2
• Music Generation: Gemini 3
• Prompt Design and Structuring: Gemini 3, ChatGPT 5.2

The dataset functions not merely as a reference, but as a system of rules (mapping, constraints, motion conditions), which are executed by AI models to generate and transform music, images, and video.

Generation Pipeline (Summary)

• Input: Selection of rokkaku_Japan-Korea_1 from the dataset + full review of exhibition texts
• Image → Music: Establishment of mapping rules (5-1), followed by generation of ambient music using Gemini 3
• Character Realization: Prompt-based definition of fixed elements, style, and constraints; generation of live-action imagery via Kling 3
• Multilingual Voice: Script in Korean, German, and Japanese → synthesized using ElevenLabs Multilingual v2
• Video: Integration of generated audio and motion rules (5-3) → production of a 16:9 video via Kling 3 → final MP4 output

Algorithm and Prompt Structure (Translation Rules)

5-1) Image → Music Mapping (Custom Rules)

• Brightness / Contrast: Brighter compositions emphasize high-frequency melodic ranges (2kHz–10kHz), with dynamic contrast shaping accents and background layers
• Texture (Brushwork): Rough textures translate into percussive elements and white noise; smooth color fields into synth pads with long reverb
• Density / Pattern: Dense compositions generate sixteenth-note arpeggios and ostinato patterns
• Gaze Scan: Left-to-right movement is mapped to stereo panning; top-to-bottom transitions shift from high to mid-low frequency ranges
• Atmosphere / Structure: Dreamlike ambient textures combined with subtle glitch effects, with the final segment looping back toward the beginning

5-2) Character Realization Conditions (Core)

• Fairytale-like live-action tone
• East Asian facial features
• Multicolored dyed hair
• Minimal head accessories
• Balanced composition to prevent overemphasis on the face

5-3) Video Generation Conditions (Motion Rules and Constraints)

• Character: A single consistent identity (no drift), natural lip-sync, minimal micro-movements (blinking, subtle gestures)
• Cinematography: Soft warm lighting, shallow depth of field, stable facial framing, slow 360° camera movement centered on the subject
• Background (reinterpreted impasto world) reacts to music:
  ① Pigment dust → rhythmic pulses per beat
  ② Smoke clouds → swirling with melody
  ③ Bokeh → high-frequency sparkle
  ④ Color clusters → two-beat drift
  ⑤ Glitter → brief response to vocal accents
• Prohibited: identity drift, additional characters, facial artifacts, flicker/jitter, warping/melting, text, watermarks, logos

Dataset–Model Interaction (Generative Logic)
The visual elements of rokkaku_Japan-Korea_1 are translated into musical parameters (5-1), forming a system of “sonic rules,” which are executed by Gemini 3 to generate music. Fixed conditions for character realization (5-2) and constraints for motion and video generation (5-3) are embedded within the prompt structure and executed by Kling 3 to produce images and video. Multilingual narration is integrated through the ElevenLabs pipeline, forming the narrative layer.

In conclusion, the dataset operates not as a passive reference but as a rule-based system (mapping, constraints, motion), and the execution of these rules by AI models constitutes the generative process that ultimately defines the work’s aesthetic.