[Talk Event] AI-Driven Production Workflows You Can Use Starting Tomorrow

We often hear about cases where AI tools are utilized, but I frequently hear designers say “I don't know what to do specifically” or “I'm frustrated because I only end up mastering ChatGPT”.

Therefore, we invited Takuya Kawai, CDO of SHIFT Inc., to speak at an online and offline event specialized in AI and design, titled “AI x Design: AI-Driven Production Workflows You Can Use Starting Tomorrow.”

Speaker Information

Takuya Kawai

In this article, we will provide a detailed, practice-based report on the AI tools discussed at this event and which tools are suitable for which types of production work. We highly recommend this to designers who want to start using AI tools to increase their productivity starting tomorrow.

*Information as of March 11, 2025. Please note that we are not conducting stealth marketing for any of the AI tools introduced.

The Current State and Future of AI x Design

According to last year's data, only about 9% of people in Japan are using AI for work. Among them, I feel that those who are interested in the theme of “AI x Design” are quite ambitious and forward-thinking.

I originally started my career as a graphic designer, and after going through web production and development, I became deeply involved in the world of AI with the advent of ChatGPT.

I will talk about AI utilization x design that you can practice starting tomorrow, incorporating project examples from prominent AI creators.

In particular, this session will proceed with content specialized in LP production. Focusing on notable tools such as Perplexity, Genspark, and Create, I will introduce design workflows using AI in three steps: “Knowing,” “Thinking,” and “Doing.”

There are many AI tools, and it is common to get lost in which one to use. Therefore, I divide them into three genres.

Imagination AI: Tools that generate text or summarize information using search tools. Example: ChatGPT
Creation AI: Tools that generate images or videos. Example: Midjourney
Assembly AI: Web and app slide generation tools. Example: Create

This time, I will introduce AI tools specialized for LP production. The above is a diagram showing the AI tools I will introduce this time, categorized by knowing, thinking, and doing.

Now, let's get straight to the main topic. This time, I would like to divide it into four phases for clarity.

1. The Current State and Future of AI x Design

2. Knowing Phase

3. Thinking Phase

4. Execution Phase

5. Summary and Reconfirming Goals

What AI Excels At and What It Struggles With

Currently, what AI excels at is gathering the following information (median values) and creating parts.

What AI struggles with is graphic design that is difficult to verbalize.

The Future of AI and Designers

I believe it will be possible to fully automate the generation of branding, products, services, and marketing just by inputting core prompts and purposes.

The designer's role is only 0→1 and 99→100.
The 0→1 part is creative direction.
The main role is defining and determining the path, while 99→100 is selection and final adjustment.
And I think a designer's aesthetic sense will become important.
In other words, I believe the role of a designer in the AI era will be to understand the philosophy and choose the best form.

Now, I would like to move on to a concrete AI-driven production workflow.

Discovery Phase

Goal: Make AI a secretary who knows the client inside and out
This time, we will use a tool called Felo.
I will explain how to research while describing the functional features of Felo.

1. Can perform deep and broad web searches

While a normal search might yield 7-8 articles, using deep research allows you to search from about 200 articles.

2. Can use the latest LLM models

You can use various models such as DeepSeek AI, Claude, and Gemini.

3. Can view information from a bird's-eye view using mind maps, AI diagrams, and slides

Features like NotebookLM are convenient.
*NotebookLM is a tool that allows you to feed information to an AI and customize your own AI.

4. You can gather information using Felo's web search

The accuracy of the information obtained increases by making the topic a specialist for the client.
That concludes the 'knowing' and 'gathering information' phases.

I believe the 'knowing' phase is the most important. This is because if you do not know things properly when you reach the 'thinking' or 'doing' phases, the direction of your output will be off.

Thinking Phase

Goal: Create a blueprint for an LP
I don't think every designer's goal is to create an LP, but if you abstract it, it can be applied to any media, so please take a look.

First, let's confirm the core message.
Make it a specialized topic, such as company information or regulations on how to create an LP.

Next, we will perform information design.
With Felo, besides text, you can organize and view information using mind maps, or organize it using AI diagrams.

The diagram above is the information design for the LP we are producing at our company.
It is possible to use AI to create diagrams to view and think about or organize information.

Next, we will generate prompts for image generation, which is essential for an LP.
I would like to generate prompts to create images to include in the LP.
I will continue to ask Felo to generate the prompts and summarize them in a spreadsheet.
I will send the spreadsheet containing a large amount of information to Gemini.
This is the first time I use Gemini here.

Also, when generating prompts for LP information design, I recommend making them HTML-style prompts by separating sections with <section> and the content of the section with </section>.

Doing Phase

Goal: Pass the work to the designer
Since there are several flows in the 'doing' phase, I will introduce them in Parts 1 to 3.

Part: 1 Image Generation

The tools I would like to introduce are the following two.

• Genspark

• Recraft

Especially this time, I would be happy if you could remember at least Genspark.

I will copy and paste the prompt generated earlier with Gemini into Genspark.
Then, after a few seconds, 4 to 6 images with different directions will be presented. If you feel that one of them 'looks right,' you can dig deeper from there.
Below are the images I generated using Genspark.

Looking at this, I feel like you can generate just about anything.
Since background removal is also possible, I think it will become a viable alternative to stock image services and image editing software.
That concludes the section on utilizing Genspark.

Next is Recraft.
Recraft isuseful when you have a specific style in mind or want to create a mock-up.It has a very strong personality, so I think it might be difficult for commercial designers to use, but as it evolves, I think it will become easier to use.
I would recommend it for merchandise development.

This automatically suggests merchandise using my icon.
By combining it with OpenAI's Sora, you can generate promotional videos.
You can also generate videos using tools like Hailuo AI.

Part 2: Website Generation

What I want to introduce here is the tool I mentioned at the beginning,Create.

Just paste the prompt generated by Gemini, and it will immediately generate a landing page design. From a professional perspective, there are still parts that need work, but if you use the approach of modifying the landing page generated by Create, it seems like you could reduce the workload.

I would also like to introduce a tool called Onlook.

This isa tool similar to the Figma UI that outputs designs for PC and mobile.
I can also recommend this to those who are used to Figma and want to reduce man-hours.

Part 3: Designer's Finishing Touches

Goal: AI passes the ball, and the designer takes the shot.
This is the final workflow.

This is a tool that allows you to paste preview data after a website is published into Figma.
You can design what you just generated in Figma. It's very convenient, isn't it?

I recommend converting images to solid colors.

This is because if images are included, it can take time if an error occurs during copying.
The flow is to edit it in Studio later and deliver it to the client.
Corrections and revisions are also easier to control in Studio.
If you need a new section, it's fine to just generate that part with Create.

Summary and Reconfirming the Goal

Today's goal wasto be able to use AI for design production starting tomorrow.

To summarize what we've discussed so far, you can utilize AI in this flow:

With these three steps, not only will your production speed increase, but the time spent worrying or feeling stuck will be dramatically reduced as well.

Thank you very much for listening until the end.
I hope you will try these out when working on web production starting tomorrow.

JOOi, the organizer of this event, is a designer matching service that introduces production projects to designers. Under our vision of becoming a place that designers aspire to be, we also provide useful information to designers through events and reports like this one.

We are currently actively recruiting designers, so we look forward to receiving applications from those interested in JOOi.