DBL Experiment

Gabriella Grace Christyanti / 0371915
Game Art / Creative Media / School of Design
DBL Experiment

---

TABLE OF CONTENTS 

1. Weekly Progress
2. Reflection

---

1. WEEKLY PROGRESS

WEEK 10

WEEK 11

WEEK 12

WEEK 13

WEEK 14

---

2. LECTURES

WEEK 11

Q1: Participant Code (No ID, name or personal Details)

S01_Gemini

Q2: Using my own 3D blockout as an image guide made it easier to control the final composition of the Ai

Generation compared to using text prompts alone

10/10

Without the image guide, the AI would often end up making the generated image face the wrong side of the room

or have the wrong composition and it would need several attempts to fix the whole composition. With the block

out, though there's still some refining that needs to be done, it reaches the point where I needed it to much faster.

Q2: The Ai successfully interpreted the asset breakdown that can transferred to 3D production

5/10

Its more to 50/50 with some assets being quite similar to what was envisioned in the image generated, but some

break downs look quite different to what was originally generated. The break down also feels not complete

enough to be able to be nicely transferred to a 3D production

Q3: What is the biggest problem you faced today when trying to make Gen Al output match your concept idea

based on 3D layout given

There are certain instructions that I need to rephrase or repeat a few times until the AI finally understands what I

want. The AI generated image also sometimes stray away from what I wanted it to look like in the block out

WEEK 12

Q1 [Participant Tracking Code]: Enter your unique tracking code: [ Short Answer ] (No names, no student IDs)

S01_Gemini

Q2 [Likert Scale 1-5]: Using AI to generate close-up material probes helped me understand the tactile surface

properties of my environment design much faster than manual texturing searches.

5/5

Q3 [Likert Scale 1-5]: It was difficult to align the aesthetic style of the AI-generated close-up micro textures with the

overall visual look of my polished digital painting.

5/5 --- difficult because it needed more detail in the prompts to somewhat understand the vision and the details are

often blurry or inconsistent

Q4 [Likert Scale 1-5]: Constructing this callout sheet forced me to rethink and change parts of my initial design

layout to make sure it was physically logical and scalable.

3/5

Q5 [Open Text]: What specific micro detail or material texture did the AI help you resolve today that was not

clearly defined in your main illustration last week?

The Architectural texture which shows multiple variations of glyph blocks and panel detail patternS

WEEK 13

Q1 [Participant Tracking Code]: Enter your unique tracking code: [ Short Answer ] (No names, no student IDs)

S01_TrimSheet_Layout

Q2 [Likert Scale 1-5]: Deconstructing my environment assets into a modular trim sheet layout first gave me a much

clearer understanding of how to prompt the AI for a realistic, functional keyshot.

3 - the results varies from different AI versions

Q3 [Likert Scale 1-5]: The cinematic keyshots generated by the AI accurately maintained the structural logic and

material rules I established in my trim sheet breakdowns.

4 - the base of each material is clear and consistent enough but the shape of the environment changed

Q4 [Likert Scale 1-5]: Using the AI as a co-creator to visualize the refined narrative (smoke, lighting shifts, damage)

was faster and more effective than manually rendering atmospheric effects.

4 - it’s good but AI won’t be able to do it perfectly so you will need to make your own refinements after that

Q5 [Open Text]: When you looked at the AI-generated keyshot, did you notice any areas where the AI broke the structural or modular rules you mapped out in your trim sheet? How did you address that visual disconnect?

Yes. The AI originally added carving details that did not follow the modular trim sheet, making some assets feel

inconsistent. This was corrected by ensuring walls, floors, arches, and props reused the same trim sheet textures, while

keeping only the statue and narrative effect

WEEK 14

Q1 [Participant Tracking Code Verification]: Enter your unique tracking code:

S01

SECTION A: TESTING THE 6 CORE FUNDAMENTALS (LIKERT SCALE 1-5)

Q2 [Value Studies & Lighting]: The GenAI workflow accelerated my ability to balance macro values and complex

dual-tone lighting environments much faster than traditional rendering.

4/5

Q3 [Perspective & Proportion]: The AI introduced serious perspective distortions and scaling conflicts, requiring

significant manual drawing layers to re-establish spatial logic.

4/5

Q4 [Composition & Design Principles]: Maintaining a uniform modular rhythm and consistent thematic design

language across separate props was highly challenging due to the AI's tendency to generate random visual noise.

2/5

SECTION B: DIAGNOSING THE STRUCTURAL PROMPTING HURDLES (LIKERT SCALE 1-5)

Q5. The biggest linguistic hurdle I faced was that descriptive keywords focusing on thematic theme look (e.g.,

"industrial cyberpunk, gritty high-tech") frequently corrupted the structural functionality and geometric alignment of

the props.

1/5

Q6. To make the AI respect the structural requirements and interlocking rules of my modular kit shell, I had to stop

using art buzzwords and adapt my prompts to include highly technical architectural phrasing

2/5

SECTION C: EMPIRICAL LEARNING OUTCOME DIAGNOSTIC

Q7 [Multiple Choice]: Overall, considering the structural components, design language constraints, and the 6 checkpoints, did this integrated co-creative loop function as a successful training accelerator or a pedagogical roadblock?

4/5 Pedagogical Acceleration: It allowed me to rapidly internalize value, texturing, lighting, and thematic detail

synthesis, making my workflow highly effective.

SECTION D: QUALITATIVE EVALUATION

Q8 [Open Text]: Contrast your operational experience in Project 1 (Manual Exterior) versus Project 2 (AI-Co-Creative Interior). Did the introduction of the AI genuinely help you understand and practice the structural logic of a professional design language pipeline more effectively, or did it feel like a mechanical failure that disrupted your authorial control? Explicitly document the exact text prompting hurdles you encountered when trying to establish theme logic and mechanical asset functionality

The introduction of AI in Project 2 helped me better understand the professional design language pipeline by acting as a reference rather than replacing the design process. Compared to doing Manual Exterior, AI sped up the ideation stage by generating detailed concepts and callout-style references that helped me refine my own design direction. However, I often struggled with prompt writing because AI could not always interpret my ideas accurately, requiring repeated prompt revisions or manual adjustments to achieve the desired outcome.

Although AI produced visually detailed results, it lacked an understanding of environmental storytelling and gameplay functionality. Some generated environments contained unnecessary visual clutter, unrealistic layouts, narrow pathways, or obstacles that could negatively affect player movement. Overall, AI was a valuable tool for accelerating concept development, but the final design still depended on human judgement to refine the narrative, functionality, and practicality of the environment.