top of page
SARAH A. SCHMIDT
Art Director & Copywriter
AI Narrative Designer
Specializing in concept development, visual storytelling,
and AI-generated imagery
GYRO, JWT, DDB
Creative Systems
A collection of system-based explorations examining how visual, narrative, and AI-driven variables shape perception, behavior, and output.
Including explorations in generative models, conversational systems, and AI-driven outputs.
⸻
Color Impact System
A visual system examining how color intensity alters emotional perception within identical compositions
⸻
System Logic
Color does not just decorate a scene—it defines how it is perceived.
By holding composition constant and shifting only color intensity, this system isolates how saturation, contrast, and palette influence emotional interpretation.
System Output
⸻
Process
• Consistent composition across all outputs
• Color grading variation (muted vs saturated palettes)
• Comparative evaluation of emotional response
⸻
Outcome
A repeatable system for testing how color alone influences perception, tone, and emotional response across identical visual structures
⸻
System Capabilities
• Isolates color as a controlled variable
• Produces comparable outputs for evaluation
• Enables scalable visual testing across palettes
⸻
Identity Fragmentation System
This system explores both fragmentation and continuity—revealing how identity shifts while remaining perceptually connected
⸻
System Logic
Identity is not fixed—it is constructed through repeated visual signals and continuously reshaped by context.
By introducing controlled variation in expression, tone, and composition, this system explores how identity can fragment while remaining perceptually connected.
.jpeg)
.jpeg)
⸻
Process
• Generation of a single subject across multiple emotional states
• Controlled variation in micro-expression and tone
• Consistency in subject structure and visual identity
• Evaluation of continuity, fragmentation, and recognition
• Identity persists not as a fixed state, but as a pattern of recognizable signals across variation
⸻
Outcome
A repeatable system for generating distinct yet connected versions of identity across outputs
⸻
System Capabilities
• Maintains subject consistency across variations
• Expands emotional range through controlled inputs
• Produces scalable identity systems across formats
⸻
Conversational Tone System
A system exploring how tone functions as a controllable variable within AI-generated responses
⸻
System Logic
Tone is not subjective—it can be structured, tested, and applied systematically.
By holding informational content constant and varying tone, this system examines how shifts in language alter perception, emotional response, and user experience.
⸻
Process
• Consistent input across tone variations
• Structured shifts in tone (neutral, supportive, creative)
• Evaluation of clarity, perception, and interaction
⸻
Outcome
A scalable system for generating consistent, human-centered responses across different tones and conversational contexts
⸻
System Capabilities
• Adapts tone based on user intent
• Maintains clarity across variations
• Produces consistent, usable responses
• Scales across conversational use cases
⸻
Generative Consistency System (AI)
A system examining how generative models produce variation, consistency, and controlled drift from identical and modified prompts
⸻
System Logic
Generative models do not produce a single fixed output—they generate a range of plausible interpretations shaped by prompt structure, training data, and internal randomness.
By holding core inputs constant and introducing controlled variation, this system reveals how outputs remain stable—or begin to drift—as conditions evolve.
Consistent Outputs
Prompt Variation
Edge Conditions
⸻
Process
• Consistent base prompt across multiple generations
• Controlled variation in descriptive modifiers
• Exploration of output consistency and drift
• Comparative evaluation of model interpretation
⸻
Outcome
Demonstrates that generative systems produce ranges of plausible outputs rather than fixed results, with consistency shaped by prompt structure and model behavior
⸻
System Capabilities
• Explores how models maintain or lose consistency across repeated and modified prompts
• Reveals patterns of variation, drift, and interpretation within controlled conditions
⸻
Model Failure System (AI)
A system exploring how generative models behave under ambiguity, contradiction, and extreme conditions
⸻
System Logic
Failures are not exceptions—they are signals of how a model resolves uncertainty.
When prompts introduce ambiguity or conflict, generative systems prioritize plausibility over structural or logical accuracy, revealing the boundaries of interpretation.
⸻
Process
• Introduction of ambiguous and conflicting prompts
• Observation of distortion, misalignment, and breakdown
• Identification of recurring failure patterns
• Evaluation of how models resolve uncertainty
⸻
Outcome
Reveals that model failures are systematic rather than random, exposing how AI prioritizes plausibility over accuracy under uncertainty
⸻
System Capabilities
• Identifies how models respond to ambiguity and conflicting inputs
• Surfaces recurring failure patterns across structure, logic, and visual coherence
⸻
Object Segmentation System (AI)
A system examining how AI identifies and isolates specific objects within complex, real-world environments
⸻
System Logic
AI perception does not operate in isolation—it functions within scenes defined by overlap, similarity, and visual noise.
By introducing competing objects and environmental complexity, this system evaluates how models distinguish targets when they are no longer the most visually dominant elements.
⸻
Process
• Selection of target objects within multi-object environments
• Introduction of competing forms and visual similarity
• Variation in scale, overlap, and background complexity
• Evaluation of selection accuracy and consistency
⸻
Outcome
Shows that object recognition depends on distinction rather than visibility, with performance influenced by context and competing elements
⸻
System Capabilities
• Evaluates object identification within complex, multi-object environments
• Tests recognition under conditions of overlap, similarity, and visual noise
⸻
Object Segmentation System - NEW paste HERE
A system exploring how tone functions as a controllable variable within AI-generated responses
⸻
System Logic
Tone is not subjective—it can be structured, tested, and applied systematically.
By holding informational content constant and varying tone, this system examines how shifts in language alter perception, emotional response, and user experience.
⸻
Process
• Consistent input across tone variations
• Structured shifts in tone (neutral, supportive, creative)
• Evaluation of clarity, perception, and interaction
⸻
Outcome
A scalable system for generating consistent, human-centered responses across different tones and conversational contexts
⸻
System Capabilities
• Adapts tone based on user intent
• Maintains clarity across variations
• Produces consistent, usable responses
• Scales across conversational use cases
⸻
Segmentation Failure System (AI)
A system exploring how object segmentation breaks down under occlusion, low contrast, and visual similarity
⸻
System Logic
Segmentation systems rely on visual boundaries, contrast, and recognizable structure.
When these signals become unclear—through overlap, obstruction, or similarity—models produce incomplete, incorrect, or unstable selections.
⸻
Process
• Introduction of occlusion and overlapping objects
• Reduction of contrast between target and surroundings
• Increased similarity between object categories
• Observation of incomplete or incorrect selections
⸻
Outcome
Highlights the limitations of visual perception models, revealing gaps between detection and true object understanding
⸻
System Capabilities
• Examines segmentation breakdown under occlusion and low contrast
• Identifies instability in boundary detection and object grouping
⸻
AI Creative Workflow System
A system outlining how AI is integrated into the creative process as a tool for exploration, iteration, and execution.
⸻
System Logic
AI functions not as a shortcut, but as a generative partner within a structured creative system.
It enables rapid exploration of directions, expansion of ideas, and iterative refinement—while maintaining concept and intent as the primary drivers.
⸻
Process
• Prompt → Generate → Evaluate → Refine
• Use of AI tools for ideation and prototyping
• Iterative refinement across outputs
• Selection and development based on concept strength
• Integration into broader creative workflows
⸻
Outcome
A repeatable workflow that produces stronger, more intentional creative results over time
⸻
System Capabilities
• Improves output through iteration
• Maintains concept integrity across cycles
• Scales across visual and narrative formats
⸻
These systems reflect an ongoing exploration of how creative direction, structured inputs, and AI models interact to shape outputs and user experience.
⸻
bottom of page