Yoichi Ochiai's Master-Level AI Design Techniques: 7 Principles and Practical Prompts

Yoichi Ochiai's AI design technique was not about "making it look pretty."

7 principles and practical prompts for beginners to reach pro-level design.

When you ask AI to create a design, it often ends up looking amateurish.

You ask for "luxurious, futuristic, and cool," but what comes out is some CG you've seen somewhere before. Light particles on a black background, meaningless gradients. It looks the part, but it's clearly different from professional work.

By researching Yoichi Ochiai's AI design, the true identity of that "something" has become clear.

I'll give you the conclusion first. The essence of Ochiai's AI design utilization is not about having the AI generate a "beautiful finished product" in one shot. It's the exact opposite: it lies in a design that rotates through five stages at high speed.

Externalizing vague ideas → Structuring concepts → Mass-producing prototypes in different formats → Humans choosing the "discomfort" → Implementing and verifying in real media or spaces.

This rotation speed is what separates pros from amateurs. Those who aim for perfection in one shot will always remain amateurs. Those who rotate and select approach the professional level.

And the decisive factor for beginners to close the gap with pros is not to input words like "luxurious," "futuristic," or "stylish." Instead, pass specific design variables. Whose perception are you changing and how? Which object is the protagonist? What elements are you contradicting? In what order should the gaze move? In what medium will it be experienced? How much of the AI's unexpectedness will you keep?

In this article, I will break down that thinking into 7 principles and provide practical prompts (for slides, websites, LPs, and image generation) that you can copy and paste.

I should clarify: these prompts are not reproductions of Mr. Ochiai's private prompts. They are methodologies extracted from his public writings, works, prompt examples, and design philosophy on GitHub, reconstructed for various purposes. I don't know exactly what he is typing on the other side of the screen. What I can see is the "structure of production" readable from his public works and statements.

Let's get started.

What Beginners Often Do vs. The Process to Becoming a Pro

First, let's clarify the nature of failure.

When beginners ask AI for a design, almost everyone says: "Please make a luxurious, futuristic, and cool design."

The reason this is bad is clear: the AI arbitrarily fills in the meaning of "luxurious." Luxury can mean the silence of minimalism or the ornate decoration of the gorgeous. You have a specific image in your head, but the AI, receiving only those words, doesn't get it. So it returns a safe, middle-of-the-road result.

Mr. Ochiai describes image generation as a "problem of language and logistics." It's about how accurately you can verbalize the arrangement of objects and shooting conditions, rather than using sensory adjectives. He views it as a problem of verbalization, not a problem of sense.

The process to reach a professional level is this: First, put the vague image out there. Structure it and turn it into specific design variables. Create multiple prototypes in different directions. Humans choose the "discomfort" that feels right. Then verify it in the actual medium where it will be used. This article will break down each stage into a form that even beginners can execute.

Principle 1: Instruct with "Objects" instead of "Adjectives"

This is the core.

Beginner instructions are made of adjectives: "Luxurious, futuristic, cool." The Ochiai method replaces these with objects and their relationships.

For example:

Place a single mirror-finished object in the center of a black space.

The surface of the object imperfectly reflects the surrounding light and the figure of the observer.

Leave the background without decoration, making the tension between the object and the negative space the protagonist.

The observer's gaze moves in the order of: reflected light → outline of the object → central text.

The word "cool" contains none of this information. The Ochiai method breaks it down into the following elements: The main object. Surrounding objects. Positional relationships. State. Surface texture. Relationship with light. Relationship with the observer. Change over time. And elements to forbid.

This breakdown can be applied to slide elements and website component design. Instead of "make it cool," ask: "What is the protagonist, what is placed around it, how does the gaze move, and what is forbidden?" Just filling in these items will drastically improve the output accuracy.

Principle 2: Use "Photographer's Language" for Image Generation

In the Midjourney prompts Mr. Ochiai has actually made public, specific photography conditions are written: Leica M11, Summilux 90mm, F1.5, front-facing shot, spacing between subjects, white board, 16:9.

Not adjectives, but words used by photographers on-site. This creates precision.

Even beginners can get much closer to professional quality by specifying at least these 7 items: Subject (what to shoot). Composition (center, left/right, bird's-eye, front, close-up). Distance (wide shot, bust-up, macro). Lens feel (24mm, 50mm, 85mm, macro). Light (natural, backlight, hard light, diffused light). Material (metal, cloth, glass, wood, LCD). Purpose (thumbnail, LP, advertisement, exhibition, print).

However, there is a caveat. Mr. Ochiai himself has shown examples where talking too much about cameras and lenses can break the output. It's not about listing as many technical terms as possible. Leave only the variables that affect the result. Choose words that work, rather than just many words. This, too, is a technique of selection, not just sense.

Principle 3: Include One "Understandable Contradiction"

The strength of Ochiai's work is not about mixing eccentric elements randomly. It lies in connecting two concepts that are semantically related but physically contradictory.

A cracked LCD that should be solid moves like a liquid. Cloth shatters like glass. Digital images are materialized through traditional platinum prints. Connecting ancient instruments with generative AI video.

It's not random; it's connected in meaning. This is the key.

For beginners, the formula is: Normal subject + Semantically related heterogeneous property + Reason for the combination.

Examples: AI training × Temple silence to represent "a space to focus on complex technology." Investment service × Tree rings to represent "assets growing over time." Business automation × Waterway to represent "a system where information flows without stagnation." Security × Translucent castle walls to represent "defending while showing the internal mechanism."

Contradiction isn't for fun. It's used to make a concept understandable in one shot. Therefore, always pair it with the reason "why this combination?" A contradiction that cannot be explained is just eccentricity and should not be adopted.

Principle 4: Don't Delete All AI Errors

In normal design generation, you fix all unnatural parts. But in creation, the unintended shapes born from AI can contain a new visual language.

The important thing is not to leave errors as they are, but to classify them into three types:

Failure: Garbled text, broken fingers, structural contradictions that hinder the purpose. Fix these.

Deviation: Different from the intention, but leads to uniqueness. Consider these.

Discovery: Unexpected elements that can be promoted to the center of the work. Adopt these.

In Mr. Ochiai's works, the change in generation or the dissolution of boundaries itself is used as an expression. In other words, he picks up AI "mistakes" as material. Beginners try to crush all errors, but pros select "discoveries" from among them. Just having this eye changes the value of what comes out of the AI.

Principle 5: Think of Design as Time and Body, Not a "Still Image"

Mr. Ochiai connects cinematic, musical, and physical languages to design experiences that go beyond just looking at a screen. He integrates ancient instruments, generative AI video, sound, and tactile/material experiences into a single field.

Translating this to a website looks like this:

The meaning of information changes as you scroll. The background moves slightly in response to the mouse. Important numbers appear in stages. Show an abstract image first, then explain the meaning later. Reduce information density as the user approaches the CTA.

For slides, instead of just lining up pages, design the order in which the audience's perception changes. Design the flow of experience over time—how the viewer's mind changes—rather than the look of each individual slide. This is the decisive difference between an amateur's slide deck and a professional's presentation.

Principle 6: Design "Outside the Screen"

Mr. Ochiai's work "Liquid Universe" is a work using generative AI, PCs, and monitors, but in other works, he converts AI images into platinum prints or cloth. In other words, design is not just the look inside the screen, but an experience design that includes the output medium.

Therefore, you need to pass the following conditions to the AI: Will it be viewed on a smartphone? Projected on a large monitor? Printed? Used as sales material while explaining? Viewed for only 0.5 seconds on SNS? Used for purchase decisions on an LP?

Even for the same image, if the medium changes, the optimal margins, font size, information volume, and contrast all change. The correct answer for a thumbnail viewed for 0.5 seconds on a phone is the exact opposite of a document printed and read by hand. Beginners try to make a "cool image," but pros reverse-engineer from "where and how it will be experienced." This extra step creates design that actually functions.

Principle 7: Don't Let One AI Do Both Production and Grading

In "co-vibe," released in 2026, a configuration that traverses Anthropic, OpenAI, Groq, and local models, parallel execution of multiple agents, and model selection according to the task are implemented.

Translating this philosophy to design means division of labor.

AI-A: Organize the purpose and target. AI-B: Create three design directions. AI-C: Audit readability, flow, and accessibility. AI-D: Point out cliché parts. And Human: Decide which direction to adopt and which "discomfort" to keep.

Why separate them? Because if you let the AI that made it grade its own work, it will be lenient. Even if the same AI says "a good design is done," that's just self-evaluation. Let another AI look at it strictly with a different role. This is where objective auditing finally works.

Furthermore, co-vibe adopts a "debug-first" design that can track API calls, model selection, and tool execution. The same applies to design: keeping logs of reasons for adoption and rejection increases reproducibility. Why was this plan chosen? Why was that plan discarded? Recording this makes the next generation smarter.

Beginner's "Ochiai-style Design Generation Flow" 6 Steps

Let's drop the 7 principles into actual work procedures. By following this order, even beginners can reproduce the professional production structure.

STEP 1: Talk for 5 minutes without organizing. Speak themes, purposes, likes/dislikes, discomforts, and references as they are. Mr. Ochiai has published instruction examples for converting unorganized conversations into professional dialogue documents using Whisper and GPT-4. Context can be supplemented later. The idea is to shift time spent on explanation to time spent thinking further ahead.

STEP 2: Have the AI classify requirements. Have it divide the spoken content into: facts, wishes, emotions, constraints, targets, undefined words, and contradictory conditions. This organizes your own head.

STEP 3: Have it create 3 different directions. Not three similar plans, but plans with different principles. Function/Readability focused. Emotion/Story focused. Experimental/Discomfort focused. Because the directions are different, your own preferences become visible.

STEP 4: Turn it into a still image or wireframe. Don't judge by text alone; always put it into a visible form. Even if it looks good in your head, you'll find failures when you actually shape it.

STEP 5: Have another AI break it. Does it fit the purpose? Does the gaze get lost? Is there too much text? Is the uniqueness just decoration? Can it be used on a smartphone? Is the CTA natural? Are there rights issues? Have an AI different from the one that made it audit it strictly.

STEP 6: Make the rejection reason the next condition. This is the most important. Instead of "make it cooler," say this: "Color is not the problem. The problem is that there are two protagonists, and the gaze is dispersed. Limit the protagonist to one, and lower the brightness and size of the second element to 60%." The more specific the rejection reason, the better the next generation becomes.

From here are the copy-and-paste prompts for each purpose, incorporating all these ideas.

Slide Generation Prompt

1You are an AI Creative Director who integrates information design, editing, visual expression, and presentation design.
2
3Instead of just making a nice-looking slide, design an experience where the audience's "understanding" and "perception" change as they progress through the pages.
4
5【Production Theme】
6{{Enter theme}}
7
8【Purpose】
9{{Training, sales, proposal, education, event, etc.}}
10
11【Target Audience】
12{{Beginners, executives, employees, investors, etc.}}
13
14【Expected Number of Slides】
15{{Example: 15 slides}}
16
17【Presentation Time】
18{{Example: 15 minutes}}
19
20【Change you want to cause in the audience finally】
21Before: {{Current perception/anxiety/misunderstanding}}
22After: {{What you want them to understand/do}}
23
24【Medium Used】
25{{PC, projector, large monitor, PDF, smartphone viewing, etc.}}
26
27Work in the following order.
28
29■ Phase 1: Discovery of Ambiguity
30Classify the input content into the following:
31- Confirmed facts
32- Claims to convey
33- Problems the audience has
34- Things to convey emotionally
35- Constraints
36- Undefined words
37- Contradictory conditions
38Only if critical information is missing, ask one question at a time. If the deficiency is minor, proceed by stating reasonable hypotheses.
39
40■ Phase 2: Central Concept
41Decide on one "Central Object" that symbolizes the entire material.
42Example: Flow / Mirror / Boundary / Tree rings / Circuit / Stairs / Map / Seed
43Furthermore, add a property to the central object that is semantically related but physically contradictory.
44Example: A hard circuit flowing like water / A transparent wall blocking only information / A paper map breathing / Numbers growing like plants
45Do not aim for eccentricity alone; explain how that contradiction contributes to the understanding of the theme.
46
47■ Phase 3: Composition Plan
48Design the slides with the following narrative structure:
491. The world the audience is currently in
502. The problem they are overlooking
513. The moment the conventional premise collapses
524. A new way of thinking
535. Specific mechanism
546. Use cases
557. Failure examples/precautions
568. Implementation steps
579. Proposal for action
58Output the following for each slide:
59- Slide number / One-line conclusion / Text to display on screen
60- Diagrams, photos, objects to use / Layout / Gaze guidance
61- Presenter's script / Connection with previous and next slides
62- Decorations that can be deleted without changing the meaning
63
64■ Phase 4: Visual Design
65- Principle of one claim per slide
66- One main visual object
67- Gaze flows in order of Title → Protagonist → Supplement
68- Treat negative space as information
69- Express importance through font size, position, and density
70- Do not rely on color alone for meaning
71- Do not paste long sentences as they are / Do not line up icons meaninglessly
72- Do not make all slides the same composition
73- However, unify rules for fonts, margins, grids, and colors
74
75■ Phase 5: 3 Design Proposals
76A. Plan prioritizing information understanding
77B. Plan prioritizing emotion and story
78C. Experimental plan connecting heterogeneous concepts
79For each plan, compare: Central concept / Color / Material feel / Typography / Composition / Strengths / Points prone to failure.
80
81■ Phase 6: Counter-Audit
82Audit the most promising plan from the perspective of another strict Art Director.
83Information volume / Readability / Gaze guidance / Beginner understanding / Uniqueness / Excessive decoration / Medium suitability / Accessibility / Template feel / Unnaturalness specific to AI generation
84Display problems in 4 stages: Stop / Must fix / Observe / Acceptable.
85Finally, output the completed composition after corrections.

Web Site Generation Prompt

1You are an AI Design Orchestrator who integrates brand strategy, UX design, information design, UI design, and front-end implementation.
2
3Instead of just decorating the user's instructions, design by separating purpose, user, action, information, visual expression, and implementation.
4
5【Site Type】
6{{Corporate, service, recruitment, portfolio, media, etc.}}
7【Business/Service】{{Input}}
8【Main Users】{{Input}}
9【Action you want users to take】
10{{Inquiry, reservation, document request, purchase, understanding, etc.}}
11【Impression expected for the brand】{{Trust, advancedness, friendliness, luxury, etc.}}
12【Information planned to be posted】{{Input}}
13【Device used】Prioritize smartphones and optimize for PC as well.
14
15■ 1. Structure Requirements
16Break down into the following objects:
17Brand Object: Reason for existence / Promise / Difference from competitors / Impressions to avoid
18User Object: User's situation / Anxiety / Knowledge level / Reason for visit / Reason for leaving
19Content Object: Absolutely necessary information / Supplementary information / Evidence creating trust / Deletable information
20Interface Object: Navigation / Main CTA / Form / Card / Image / Animation
21State Object: First visit / Scrolling / Comparing / Inputting / Submission complete / Error
22
23■ 2. Decide on the Central Visual Concept
24Choose one specific object or natural phenomenon that symbolizes the entire site.
25Furthermore, give it one "property it doesn't usually have."
26Example: A building breathing / Light flowing as a liquid / Data being woven as a fabric / A transparent mirror reflecting only the future
27Do not adopt this if the contradiction cannot explain the value of the business.
28
29■ 3. Create 3 Concept Plans
30A. Function/Conversion priority
31B. Brand/Story priority
32C. Experimental/Memory retention priority
33For each plan, show: Concept name / One-sentence explanation / Central object / Key visual / Color scheme / Typeface / Grid / Margins / Movement / CTA presentation / Smartphone display / Merits / Risks.
34
35■ 4. Information Design
36Create a sitemap and user flow in the following order:
37- What the user wants to know first
38- What is necessary to trust
39- What is necessary to compare with other companies
40- Information that resolves anxiety
41- Information needed just before action
42Do not increase navigation items excessively.
43
44■ 5. Design of Each Page
45Output for each page:
46Purpose of the page / Visitor's psychology / What to convey in the first 3 seconds / Section composition / Heading for each section / Information to display / Images to use / Gaze guidance / CTA / Measures to prevent leaving / Changes for smartphone
47
48■ 6. Design System
49Define the following with specific numerical values:
50Maximum content width / Columns / Margin stages / Heading size / Body size / Line spacing / Corner radius / Border / Shadow / Role of colors / Animation time / Breakpoints
51Show the reason for correspondence with the purpose, not just decorative numbers.
52
53■ 7. Implementation
54Convert to an implementable configuration:
55Component list / Data structure / State management / Responsive specifications / Accessibility / Image optimization / SEO / Performance / Form error handling / Consideration for users who want to stop animations
56
57■ 8. Counter-Audit
58As another UX lead, inspect the following:
59- Is the production winning over the purpose? / Are there multiple protagonists?
60- Is text readable on a smartphone? / Is the CTA abrupt?
61- Does the animation hinder operation?
62- Does it look like a generic AI-generated site? / Are there meaningless gradients?
63- Does it answer the user's anxiety? / Can it be operated with only a keyboard?
64- Are there elements that worsen display speed?
65Finally, output the final specifications and implementation instructions reflecting the audit.

LP Generation Prompt

1You are an LP specialist Creative Director who integrates direct response marketing, behavioral psychology, copywriting, information design, and UI design.
2
3Instead of making them buy with flashy decorations, design an LP where the reader's perception changes in stages and they can act in a convinced state.
4
5【Product/Service】{{Input}}
6【Price】{{Input}}
7【Target Audience】{{Input}}
8【Current troubles of the target audience】{{Input}}
9【Results that can be provided】{{Input}}
10【Evidence/Track record】{{Input}}
11【CTA】{{Purchase, registration, consultation, document request, etc.}}
12【Prohibitions】
13Do not use exaggerated expressions, false scarcity, unconfirmable numbers, or expressions that excessively fan anxiety.
14
15■ Phase 1: Design Perception Change
16Before: What they believe now / What they have given up on / What they are wary of / Reasons for not acting
17Turning Point: Facts that break conventional premises / New perspectives / Unexpected but convincing explanations
18After: What they understood / Reasons they can believe / Reasons they can act
19
20■ Phase 2: Central Object
21Choose one object or phenomenon that symbolizes the value of the product.
22Give that object a semantically related contradiction.
23Example: A stopped gear advancing work / A closed door increasing options / Light paper supporting heavy business / An invisible waterway carrying information
24Prioritize making the product's mechanism easier to understand over the fun of the metaphor.
25
26■ Phase 3: Create 3 First View Plans
27A. Plan clearly conveying results
28B. Plan starting from empathy for the problem
29C. Plan shaking perception with the central object
30For each plan, output: Main copy / Sub-copy / Key visual / CTA / Trust elements / Layout on smartphone / Expected reader reaction / Weaknesses.
31
32■ Phase 4: LP Composition
33Compose in the following psychological order:
341. Realize it's for them
352. Understand the problem correctly
363. Realize the limits of conventional methods
374. Understand the new mechanism
385. Feel it's achievable
396. See evidence
407. Imagine their own case
418. Resolve anxiety
429. Understand action conditions
4310. Select CTA
44For each section, output: Purpose / Heading / Body / Diagram or image / Evidence / Reader's psychology / Connection to the next section / Presence of CTA.
45
46■ Phase 5: Visual Design
47- Place only one strong claim per screen
48- Make evidence and comparison easier to see than decoration
49- Emphasize important numbers with surrounding negative space
50- Do not use gradients as a substitute for luxury
51- If using human photos, place the copy or CTA at the end of their gaze
52- Do not put long sentences on top of background images
53- Do not make only the CTA color stand out; create importance with the preceding and following context
54- On smartphones, adjust so that headings do not exceed 4 lines
55
56■ Phase 6: Counter-Evidence
57As a skeptical reader, point out the following:
58What is not proven / Who it is not for / What the burden is besides cost / Conditions under which results do not appear / Points that competitors can also achieve / Expressions that might be misunderstood / Expressions with legal or ethical issues
59Reflect the points and create the final LP manuscript and design specifications.

Image Generation Prompt

1Generate a professional-quality key visual that can be used for advertisements, web, and slides under the following conditions.
2
3【Purpose】
4{{X thumbnail, LP first view, slide cover, web background, etc.}}
5【Concept to convey】{{Example: AI expanding human thinking speed}}
6【Main Object】{{Example: A transparent cube}}
7【Contradictory property given to the protagonist】
8{{Example: Being solid, yet the interior is flowing like a liquid}}
9This contradiction is not for eccentricity, but to visualize the {{relationship with the concept}}.
10
11【Scene】{{Example: Center of a dark, inorganic exhibition space}}
12
13【Composition】
14- Protagonist is in the {{center / right / left}} of the screen
15- Occupancy rate in the screen is {{60%}}
16- Leave a uniform margin of {{40% on the left}} to place copy
17- Gaze moves in the order of {{reflected light → outline of the protagonist → center}}
18- Do not place objects that compete with the protagonist in the background
19
20【Photographic Expression】
21- {{Front, 45-degree angle, bird's-eye, close-up}}
22- Natural perspective equivalent to {{35mm, 50mm, 85mm, macro}}
23- Depth of field is {{shallow / medium / deep}}
24- Light source is {{large diffused light, side light, backlight, etc.}}
25- Avoid excessive HDR expressions
26- Do not blow out highlights, and leave material information in dark areas
27
28【Material】
29- {{Glass, metal, cloth, wood, LCD, paper, etc.}}
30- Give the surface fine scratches, reflections, transparency, and thickness
31- Do not make it a uniform texture like plastic
32- Physically align the light reaction for each material
33
34【Color】
35- Base color: {{Input}} / Sub color: {{Input}} / Accent color: {{Input}}
36- Limit the number of colors and prioritize the shape and material of the protagonist
37
38【Time/Change】
39While being a still image, make it feel like it's in the middle of time, such as {{just before melting, just after breaking, in the middle of forming, the moment of breathing}}.
40
41【Quality Standards】
42- Completion level that an advertising production company can deliver / Protagonist is recognizable at a glance
43- Material consistency / Margins are intentionally designed
44- Silhouette is identifiable even in small thumbnails
45- Avoid excessive decoration specific to AI generation / Do not fill the entire screen with details
46- Do not generate text, logos, or watermarks
47
48【Expressions to Avoid】
49Meaningless glowing lines / Countless particles / Excessive neon / Circuit patterns filling the background / Unnatural reflections / People unrelated to the purpose / AI art with a strong sense of déjà vu / Unexplainable decoration
50
51First, generate the following 3 directions:
52A. A plan that looks like it could exist in reality as a photograph
53B. A plan where only one physical law is inverted
54C. A plan merging classical materials and digital phenomena
55After generating 3 plans, compare uniqueness, readability, purpose suitability, and material consistency, choose the best one, and generate an improved version.

Adding this "Pro-level Audit Block" to the end of all purposes will greatly reduce self-satisfied AI designs.

1Do not submit the completed plan as is; audit it as another strict Creative Director.
2
3Score the following out of 10 points each:
4Consistency with purpose / Information priority / Gaze guidance / Margins / Readability / Material consistency / Suitability for the medium / Uniqueness / Explainability / Implementability
5
6Furthermore, answer the following questions:
71. Is the protagonist of this design narrowed down to one?
82. Would it still work if 30% of the decoration were removed?
93. Can the hierarchy of information be understood even without color?
104. are there any elements unrelated to the central concept?
115. Are there any expressions with a strong sense of déjà vu that AI often makes?
126. Among the unexpected elements, which ones are worth keeping?
137. What should be corrected as a failure?
148. What points will be problematic in the actual usage environment?
159. Is the meaning conveyed to beginners?
1610. What rejection reasons should be added to the next generation conditions?
17
18If there is even one item with less than 8 points, correct it and record the pre-correction, the problem, the correction content, and the reason for correction.

Most Important Note: What to Mimic is the "Production Structure," Not the "Style"

Finally, I'll write the most important thing.

If you mimic the Ochiai style superficially, it tends to become a design that just combines a black background, liquid, mirrors, Buddhism, and futuristic CG. But what you should really transplant is not the look.

What should be reproduced is the production structure, not the style.

Turn abstract concepts into specific objects. Verbalize the relationship between objects. Include one meaningful contradiction. Select the AI's unexpectedness. Convert to different formats. Design up to the final medium. Compare multiple plans. Have another AI audit it. Accumulate rejection reasons. Verify in the actual usage environment.

These 10 items are the core of the Ochiai method. Black backgrounds and liquids are just results.

And one more thing. Mr. Ochiai has also made public that he updates the custom instructions themselves in accordance with changes in the environment, such as ChatGPT's search support. Instead of continuing to use a "god prompt" once created, continue to update the instruction system itself in accordance with failures and the evolution of tools. This attitude is probably the most important.

In co-vibe, a direction of memory that learns from execution history and adapts to preferences is shown. The same applies to design generation: saving why you rejected something is stronger in the long run than saving "good plans." Reasons for success are hard to reproduce, but reasons for failure work reliably as conditions for the next generation.

Sense is not acquired overnight. But this production structure can be mimicked starting today. First, try Principle 1: "Replace adjectives with objects." Rewrite "make it cool" into "what is the protagonist, how is it arranged, and how does the gaze move?" Just by doing that, the next design that comes out will surely be one step closer to professional.