Exploring Creative Direction Through Generative AI

A personal study focused on treating AI not as a tool, but as a system to bring creative visions to life.

JUMP TO

App Development

AI-generated creative companion app.

Brand Campaign

Cohesive brand identity & asset production.

Video Production

Production-oriented video workflows.

Website Building

Translating wireframes into code.

App Development

This project is a fully AI-generated application designed as a creative workspace for ideation, planning, and collaboration. It combines multiple tools into a single system, allowing users to capture thoughts, develop ideas, and share them through an integrated social feed. The app includes features such as a collaborative feed, a dedicated “companion mode” for focused creative sessions, and an AI-driven location finder that enables users to search for filming locations by describing a scene’s mood, style, or environment. It also includes full user authentication and the potential for external database integration, allowing for persistent user data and account-based functionality.

Concept Development & Feature Ideation ▼

The application concept originated as a personally directed creative project, with ChatGPT used as a collaborative tool to refine and expand the idea. The direction, goals, and overall vision were defined from the outset, with ChatGPT supporting ideation, feature exploration, and structural clarification throughout the early stages.

Design Translation & Prompt Structuring ▼

A hand-drawn wireframe was created alongside defined design principles such as color palette, typography, and intended functionality. This creative direction was provided to ChatGPT, which translated the input into a structured, AI-readable prompt suitable for implementation. This step focused on converting human design intent into a precise format for AI execution.

AI Application Generation ▼

The structured prompt was then provided to Base44, which generated the initial version of the application. The output included the core interface, system structure, and implemented features based on the directed specifications.

AI Iteration, Debugging & Refinement ▼

The application was iteratively refined through Base44’s AI-driven development environment. This included debugging errors, adjusting UI elements, and refining functionality through continuous prompt-based interaction until the application aligned with the intended creative direction.

Strong at full-stack generation

Effective at self-debugging

Inconsistent UI updates across pages

Weak at creative UI design

Bullet points > long prompts

Better at logic than design

Brand Campaign

This project explores the use of AI to simulate a full mini advertising campaign, using generative tools to create a cohesive brand identity and multi-format content system. The goal was to examine how AI can support the end-to-end creative workflow, from brand definition to asset production, while maintaining consistency across both static and motion outputs. The campaign consists of a fictitious brand developed in collaboration with AI, translated into a set of visual guidelines including logo, typography, color palette, and overall mood. These guidelines were used to generate a series of coordinated social media assets designed to function as a unified campaign.

Brand Creation & Direction ▼

A fictitious brand was developed and refined using ChatGPT, which was used as a collaborative tool to expand the concept and define the overall creative direction of the campaign.

Brand Guidelines & Style System ▼

Core brand elements such as logo direction, color palette, typography, tone, and visual mood were defined. ChatGPT was then used to translate these inputs into a structured style guide, formalizing the brand system for consistent use across all generated assets.

Workflow Automation ▼

Google Opal was used as an AI workflow automation tool to build a structured generation pipeline for the campaign. Within Opal, a system was created to store and apply brand guidelines and product images, enabling consistent asset generation without repeatedly re-entering constraints.

AI Asset Generation ▼

Campaign visuals and videos were generated in Google Opal using Google Gemini and Google Veo 3 based on my creative direction and visual ideas defined during planning. Assets were iteratively refined through repeated AI generation until they aligned with my intended vision and quality standards.

Caption Generation ▼

Social media captions were generated using AI, with natural language prompts used to guide tone, messaging, and alignment with the brand identity. These captions were paired with the generated visual assets to complete the campaign outputs.

Maintains branding across outputs

Automation reduces repetitive prompting

Text placement & spelling accuracy in video is unreliable

Static better than motion

Multiple iterations are required to obtain usable outputs

Video Production

This project explores the use of AI in a production-oriented video workflow, with the goal of creating a commercial using entirely AI-generated assets. The intent was to test how far AI could automate the video creation process, from asset generation to final editing. An initial attempt was made to fully automate editing using an AI video editor, but due to limitations in narrative control and timing, the workflow shifted to a hybrid approach. AI was used to generate all visual and audio assets, while final editing was completed manually.

Concept, Narrative & Shot Planning ▼

The commercial concept, narration, and shot structure were fully pre-planned before production, with a clear intention to execute the entire piece using AI-generated assets.

AI Asset Generation ▼

The narration was generated using ElevenLabs AI, video clips were created using Google Veo 3, and background music was generated using Suno AI. These components formed the core asset pipeline for the commercial.

AI Editing Experiment & Workflow Shift ▼

An AI video editor was tested using all generated assets to automatically construct the commercial. The output failed to meet narrative and pacing requirements, particularly in terms of structured storytelling, leading to a shift away from full automation.

Manual Editing & Narrative Assembly ▼

The commercial was fully assembled through manual editing to maintain narrative control, pacing, and creative intent, using AI-generated assets as the foundational building blocks.

Iterative Asset Regeneration ▼

Additional video clips and supporting assets were regenerated as needed during the editing process to replace or refine specific shots without requiring new production from scratch.

AI excels at generating production assets quickly

Regenerating assets replaces traditional reshoots

Story structure still requires manual direction

AI struggles with cohesive long-form storytelling

Automated edits are suitable for rough assembly only

Hybrid workflows outperform full automation

This project explores the challenge of maintaining consistency across AI-generated video clips within a narrative structure. Using a simple story centered around recurring characters and environments, the goal was to create a cohesive sequence of scenes generated entirely through AI, despite being limited to short clip durations. The focus was not on realism, but on continuity—ensuring that characters, environments, lighting, and overall narrative flow remained consistent from clip to clip. This project functions as a controlled study on how prompting, reference inputs, and constraints influence the ability of AI to produce visually and narratively coherent sequences.

Character & Environment Setup ▼

Character visuals were generated using AI, while a stock image was selected as a reference for the environment to establish a consistent visual foundation.

Prompt Translation ▼

Scene ideas were described in natural language and translated by ChatGPT into concise, structured prompts optimized for video generation.

Prompt Structure Development ▼

A repeatable prompt format was established, consisting of scene description, camera movement, and action direction to improve output consistency.

Iterative Clip Generation ▼

Clips were generated and refined through repeated iterations, adjusting prompts to better align outputs with the intended vision.

Editing & Audio Generation ▼

Clips were manually edited into a sequence, followed by AI-generated soundtrack creation based on the final video.

In prompts, include scene, camera, and action descriptions

Reference images stabilize character consistency

Last frame of one clip as a reference for the first frame of the next

More constraints improve output accuracy

Multi-clip continuity is difficult to perfect, but possible

High iteration is required for usable sequences

This is an AI product reveal focused on experimenting with creating striking motion visuals through the use of AI.

Building This Website

This project focuses on the creation of this portfolio website using AI as a development tool to translate design direction into a functional interface. The goal was to explore how AI can be used to assist in front-end development while maintaining full creative control over layout, structure, and user experience.

Wireframe Creation ▼

A basic wireframe was created using Figma to define layout and content placement, focusing on structure rather than detailed design. The wireframe reference is provided on the right. Click to expand.

Prompt Creation ▼

A detailed prompt was written outlining design direction, visual constraints, and required functionality for the site.

Initial Generation ▼

The wireframe and prompt were provided to Google AI Studio, using the wireframe as a layout reference and the prompt to guide design and implementation.

Iteration & Refinement ▼

The site was iteratively refined through prompt adjustments, improving layout, styling, and functionality over multiple iterations.

Content Integration ▼

Final content was added and adjusted within the generated structure to complete the site.

Strong at generating initial code

Changes can unintentionally change layout

Great at translating a wireframe to functioning code

Requires constant oversight and validation

Website Visual
Click to enlarge

What I Learned

Design

AI can generate visuals, but strong design still relies on human direction, taste, and intentional decision-making.

Prommpting = Control

Output quality is driven by how clearly ideas are communicated. Concise, structured prompts work better than long, unfocused ones.

Workflow

AI accelerates the creative process, speeding up mundane, repetetive tasks and allowing ideas to be executed and refined quickly while maintaining full creative control.

Where I Will Use It

I believe I will utilize AI in my workflow by using it to refine concepts, visualize them, & generate assets. I have noticed that the best results come from using AI as an assistant rather than to fully generate media.