SHINSAN DESIGN

Challenge

A short brand film expressing SHINSAN DESIGN’s approach, where AI, design and implementation work as one continuous process. Multiple shots were connected into a coherent sequence while preserving the visual language of the source images.

AI Direction

Each shot was generated separately in Seedance from a reference image. Instead of generating one long sequence, motion, camera behaviour and restrictions were specified shot by shot, then the most stable results were selected and edited.

Process

Narrative Structure → Reference Image → Shot Design → Seedance Generation → Shot Selection → Premiere Editing → Music & Transitions → Web Delivery

Evolution

Building on the earlier CASE 02, the workflow evolved toward reference-led visual direction, shot-level control and deliberate post-generation editing. The evolution of the AI video-production process is documented as part of the work itself.

Starting Point

An early experiment in bringing generative AI into video production. Generated material and editing were combined to communicate the brand direction through motion rather than words alone.

AI Workflow

Visual material was generated with Gemini, while ChatGPT and Claude were used to structure the narrative and copy. Premiere was then used for cutting, typography and direction, rebuilding the material as a brand film for the web.

Process

Concept → Structure → AI Asset Generation → Premiere Editing → Typography & Timing → Web Delivery

Learning

This early work became the starting point for CASE 01, which introduced reference images, shot-level control and short segmented generations. The portfolio records not only finished work but also how the production method itself has evolved.

Business Issue

Job listings alone often fail to communicate the work, culture and potential for growth after joining. The experience was redesigned around the candidate decision journey so visitors could picture themselves working there.

Hypothesis / Solution

Rather than presenting information in the company’s preferred order, the site follows the candidate’s questions: company values → role → workplace → onboarding journey → FAQ → application, implemented as a responsive website.

Process

Research → Candidate Journey → Information Architecture → UI Design → HTML / CSS / JavaScript → Responsive Design → Interaction Implementation

Outcome

Recruitment problem definition, design, AI use and front-end implementation were connected in one workflow, resulting in a live prototype where the full journey from browsing to application can be experienced.

Challenge

An experiment in AI-led education and training content, testing whether a presenter, device interaction and information display could work together as one coherent video.

AI Direction

Generated in Seedance from a character reference, with a focus on human performance and device interaction while separating areas where generative AI performs well from fragile areas such as text and UI.

Process

Character Design → Reference → AI Video Generation → Shot Selection → UI & Text Review → Premiere Editing → Web Delivery

Learning

Presented as a practical example of AI video direction: generated output is treated as source material, usable expressions are selected, and human editing is used to bring the work to a publishable standard.

Experiment

A 15-second experiment built from the abstract theme “SHINSAN DESIGN” and only minimal direction, leaving much of the visual narrative to the AI. It explores how the model constructs a brand image through people, data and collaboration.

What Emerged

The result produced unexpectedly strong visual ideas, but also unintended UI text, brand elements and inconsistencies. The first-generation state is deliberately shown to document what happens when more creative discretion is given to the AI.

Refinement Workflow

GENERATE → SELECT → REFINE → EDIT → DELIVER. Rather than accepting generated output as a finished product, usable shots are selected, references, regeneration or partial fixes are applied where needed, and timing, sound, transitions and typography are rebuilt in Premiere.

Learning

Where CASE 01 designs first and then controls the AI, this experiment lets the AI explore first and has a human evaluate and edit afterwards. It treats generative AI as a source-material engine and turns the human process of judgement and direction into the case study itself.