AI Image Generation and Design · AI + human design workflows · lesson 16 of 18 · 7 min

Where AI fits in the creative process

AI changes tasks, not the need for design judgment

AI image tools compress some tasks from hours to minutes — exploration, background creation, retouching, format adaptation. They do not replace the core of design work: understanding the problem, making choices, crafting details, ensuring accuracy and taking responsibility. The most effective teams redesign their workflow around this reality.

Mapping AI to the design process

| Stage | Human role | Where AI helps | Risk if AI leads | |---|---|---|---| | Understand the brief | Ask questions, define objectives | Summarize research, generate question lists | Solving the wrong problem | | Explore | Direct exploration, judge ideas | Rapid moodboards, many visual directions | Generic, trend-driven ideas | | Concept | Choose and articulate the idea | Visualize concepts quickly | Pretty images without a concept | | Develop | Craft layouts, typography, systems | Generate assets, backgrounds, variations | Inconsistency, off-brand details | | Refine | Detail, accuracy, brand fit | Inpainting, upscaling, cleanup | Artifacts, errors | | Review | Critique, compliance, ethics | Checklists, alt-text drafts (reviewed) | Missed legal or cultural problems | | Deliver & learn | Present, document, measure | Resizing, file prep | Poor documentation |

Three collaboration models

AI as sketchbook: AI is used only for ideation and moodboards; final assets are photographed, illustrated or designed by humans. Lowest risk; common in brand identity and premium work.
AI as production assistant: humans design the system and key assets; AI produces backgrounds, variations and edits within that system. Common for social content and ads.
AI as primary image source: most imagery is generated, with human direction, editing and QA. Suitable for stylized content where accuracy and authenticity aren't central; requires strong governance.

Choose the model per project based on authenticity needs, risk, budget and client policy.

Skills that matter more now

Art direction: describing and judging visual outcomes precisely.
Taste and editing: selecting the best outputs and knowing what to discard.
Systems thinking: building style systems, templates and pipelines.
Craft: typography, composition, retouching, compositing.
Domain knowledge: accuracy in products, culture, regulated industries.
Ethics and compliance literacy: rights, consent, disclosure.
Communication: explaining AI use and its limits to clients.

Time savings — and where they go

AI can save significant time in some tasks, but review, editing and QA take time too, and poorly managed iteration can waste hours. Reinvest time savings in:

Better concepts and more directions explored.
More careful QA and accessibility checks.
Testing and learning (more ad variations, better measurement).
Human craft on the elements that matter most (hero images, typography).

Worked example: a two-person creator studio

The studio produces weekly content for three clients.

Before AI: stock photos plus manual design; limited variety.
After redesign: AI moodboards for monthly planning (sketchbook); AI backgrounds and illustration within approved style systems (production assistant); real photography for products and people; a 30-minute weekly QA session.
Outcome measured: faster turnaround on routine posts, more concept options for clients, and fewer stock-photo clichés — with no AI used where authenticity mattered.

Common mistakes

Letting AI outputs define the concept instead of the brief.
Using the same collaboration model for every project.
Counting generation time but ignoring review time.
Cutting human craft on the most visible assets.

Designing your own workflow

Start small. Pick one repetitive task — for example creating backgrounds for weekly posts or extending images for vertical formats — and introduce AI there with a clear quality standard. Measure how long the task took before and after, including review time, and note any quality issues. Once the change is stable, move to the next task. This incremental approach keeps quality high, helps clients and teammates build confidence, and prevents the chaos of changing every part of the workflow at once.

Keeping people at the center

AI tools should support the people doing the work, not deskill them. Continue investing in drawing, photography, typography and composition skills, because they are what allow you to judge and improve AI outputs. Teams that keep strong craft skills can switch tools easily when the landscape changes; teams that rely entirely on one generator are exposed when its terms, pricing or quality shift.

Hands-on: map one project and decide the collaboration model

Use this worksheet at the start of every project. It takes ten minutes and prevents most "we should not have used AI for that" moments.

PROJECT AI MAP — project: __________  client policy on AI: __________
Authenticity need (does the audience rely on images as evidence?)  low / med / high
Accuracy need (products, culture, regulated claims)                low / med / high
Ownership need (logo, mascot, key visual to protect)               low / med / high
Volume and speed pressure                                          low / med / high
→ Collaboration model: sketchbook / production assistant / primary source

Stage          | Human owner | AI allowed for          | Must stay human
Brief          |             | research summary        | objectives, audience
Explore        |             | moodboards, directions  | choosing the idea
Develop        |             | backgrounds, variations | layout, type, system
Refine         |             | inpaint, upscale        | accuracy, brand fit
Review         |             | checklist drafts        | sign-off, ethics
Deliver/learn  |             | resizing, file prep     | presentation, insight

A simple rule: if any "need" is high, move one step toward the sketchbook model. A mascot with high ownership need means AI for exploration only; a product with high accuracy need means real photography plus AI context.

Before/after: a studio week

| | Before (ad hoc AI use) | After (mapped workflow) | |---|---|---| | Monday | Jump straight into generating hero images | 30-minute brief with AI questions; model chosen per asset | | Mid-week | Rework when the client rejects "generic" images | Concepts approved as rough AI boards before production | | Friday | Late QA, disclosure forgotten | QA and disclosure built into the schedule | | Time saved used for | More generations | Better concepts, client strategy, real photography where it matters |

(Illustrative, based on a typical small-studio pattern.)

Skills to practice this month

Art direction language: describe five images you admire in the seven-part structure without naming the source.
Editing and taste: from a batch of 20 outputs, choose one and write three reasons.
Critique: run a structured critique (hierarchy, type, color, accuracy, brand, ethics) on one AI-assisted layout per week.

Summary

Map AI to specific stages of the design process, choose a collaboration model per project, strengthen art direction, taste, systems and compliance skills, and reinvest time savings in concept quality and QA.

Video lecture: Where AI fits in the creative process

Lecture coming soon · 11 chapters · about 9 minutes. Read the full transcript below.

Where AI fits
Why it matters
AI across the process
Three collaboration models
Choosing a model
Worked example 1: a pet-portrait illustrator
Worked example 2: a two-person studio
Watch me do it: the project AI map
Skills that matter more
Common mistakes
Recap and try this now

Lecture transcript

Where AI fits

Here's a question I hear from designers all the time, usually a little nervously. If AI can make images in seconds, what am I for? It's a fair question. And the answer is surprisingly encouraging, as long as you understand where AI fits in the creative process and where it doesn't. AI compresses some tasks from hours to minutes. It doesn't replace understanding the problem, making choices, crafting details and taking responsibility. In this lecture, you'll map AI onto the design process, learn three collaboration models, see which skills matter more now, and use a worksheet to decide how much AI belongs in any project. By the end, you'll have a clear, confident answer to that nervous question.

Why it matters

Why does this matter? Because teams that bolt AI onto an old process tend to get the worst of both worlds. Faster generation, but more rework, generic ideas, and missed legal or cultural problems. Teams that redesign the process around AI get real gains. Think about power tools in carpentry. A nail gun doesn't make someone a carpenter. It makes a carpenter faster at one step. You still need the plans, the measurements and the finishing. And a nail gun in the wrong hands makes mistakes faster too. AI image tools are the power tools of visual design. Your job is to know which step they belong in.

AI across the process

Let's map AI to the process. Understand the brief: humans ask questions and define objectives, and AI can summarize research. The risk if AI leads? Solving the wrong problem. Explore: humans direct and judge, AI produces rapid moodboards and many directions. Risk: generic, trend-driven ideas. Concept: humans choose and articulate the idea, AI visualizes it quickly. Risk: pretty images with no concept. Develop: humans craft layouts, type and systems, AI produces backgrounds and variations. Refine: humans check detail and accuracy, AI inpaints and upscales. Review: humans own critique, compliance and ethics. And deliver: humans present and measure, AI helps with resizing and file prep. Here's the key idea. Humans own every decision. AI speeds up the making.

Three collaboration models

Now the three collaboration models. Model one, AI as sketchbook. AI is used only for ideation and moodboards, and final assets are photographed, illustrated or designed by humans. Lowest risk, and common in brand identity and premium work. Model two, AI as production assistant. Humans design the system and key assets, and AI produces backgrounds, variations and edits within that system. Very common for social content and ads. Model three, AI as primary image source. Most imagery is generated, with human direction, editing and quality control. It suits stylized content where accuracy and authenticity aren't central, and it needs strong governance. You choose the model per project, not once for your whole career.

Choosing a model

How do you choose? Use four questions. How much does the audience rely on the images as evidence? That's authenticity need. How much accuracy matters, for products, culture or regulated claims? How much the client needs to own and protect the asset? And how much volume and speed pressure there is. Here's my rule of thumb. If any of the first three needs is high, move one step toward the sketchbook model. A mascot with a high ownership need? AI for exploration only. A product with a high accuracy need? Real photography, with AI for context. A monthly batch of stylized blog illustrations with low risk and high volume? Production assistant or even primary source works well.

Worked example 1: a pet-portrait illustrator

Worked example one, simple. A solo illustrator in London who makes custom pet portraits. Should she use AI? Let's run the questions. Authenticity need: high, because customers want their own pet, painted by her. Ownership: high, because her style is her business. So AI is a sketchbook at most. She uses it to explore background ideas and color moods for each commission, then paints the portrait by hand. She tells customers exactly that on her website. AI didn't replace her. It gave her ten background options in five minutes, so she spends more time on the part customers actually pay for.

Worked example 2: a two-person studio

Worked example two, a business scenario with illustrative details. A two-person creator studio in Islamabad produces weekly content for a skincare brand and a fintech app. Before, they used AI ad hoc. They jumped straight into generating hero images, got rejected for generic results, and forgot disclosure on Fridays. So they mapped their workflow. For the skincare brand, high accuracy need, they chose production assistant: real product photography, AI backgrounds. For the fintech app, stylized illustrations and low authenticity need, they used AI as a primary source inside a strict style system. They moved concept approval earlier, with rough AI boards. And they spent the time they saved on client strategy and one real photoshoot a month.

Watch me do it: the project AI map

Watch me do it. I'll fill in the project AI map for a new client, a café chain opening its fifth branch. First, the client's AI policy: allowed for illustrations and backgrounds, never for food. I rate the needs. Authenticity: high for food, low for decorative illustrations. Accuracy: high, because menu items must look like what customers get. Ownership: medium, since there's a mascot. Volume: high. So the model is production assistant, and sketchbook for the mascot. Then I fill in the stage table. Brief, human. Explore, AI moodboards. Develop, AI backgrounds, human layouts. Refine, AI inpainting, human accuracy checks. Review, human sign-off. That's ten minutes, and everyone knows the rules.

Skills that matter more

Let's talk about the skills that matter more now. Art direction, meaning the ability to describe and judge visual outcomes precisely. Taste and editing, choosing the best output and knowing what to throw away. Systems thinking, building style systems, templates and pipelines. Craft, like typography, composition, retouching and compositing, which is where AI images become finished work. Domain knowledge, for accuracy in products, culture and regulated industries. Ethics and compliance literacy, around rights, consent and disclosure. And communication, explaining choices to clients. Notice something? These were always the valuable parts of design. AI just made them more obvious.

Common mistakes

Common mistakes. Letting AI lead the concept stage, so you get beautiful images with no idea behind them. Using one collaboration model for every project. Treating time saved as a reason to cut prices to the bone, instead of reinvesting it in better thinking. Removing human review because the images look good. Forgetting that some clients prohibit AI, and not asking. And neglecting your craft skills. If you can't retouch, set type or build a layout, you can't finish AI output properly. Keep people at the center. The audience, the client, and the people depicted in the images.

Recap and try this now

Let's recap. AI compresses tasks, not judgment. Map it onto your process so humans own every decision and AI speeds up the making. Choose a collaboration model per project, sketchbook, production assistant or primary source, based on authenticity, accuracy, ownership and volume. If any of the first three needs is high, move toward the sketchbook. And invest in the skills that matter more now: art direction, taste, systems, craft, domain knowledge, ethics and communication. Try this now. Take one recent project and fill in the project AI map from the lesson. Mark where AI helped, where it could help, which model fits, and where human craft must stay.

Key takeaways

AI compresses tasks such as exploration, backgrounds and resizing; humans own the brief, concept, decisions, accuracy and responsibility.
Choose a collaboration model per project: sketchbook, production assistant or primary source.
If authenticity, accuracy or ownership needs are high, move toward the sketchbook model.
Art direction, taste, systems thinking, craft, domain knowledge, ethics and communication matter more, not less.
Reinvest time saved into better concepts, strategy and real photography where it counts.

Try it

Map one of your recent projects onto the process table. Mark where AI helped or could help, which collaboration model fits, and where human craft must stay.