Building Mireka: A Japanese-First AI Image Platform as a Solo Developer

I recently launched Mireka.jp, an AI image generation platform designed primarily for Japanese users.

Mireka is still at an early stage. It is not a finished product, and I do not want to pretend that everything has already been figured out.

I am building it in public: launching early, observing how people use it, identifying what feels confusing, and improving the product step by step.

In this post, I want to share why I started Mireka, why I chose to focus on Japan first, what I learned after launching the first version, and how I plan to expand it into a multilingual product.

The Problem I Wanted to Solve

There are already many powerful AI image models available today.

Some are better at photorealistic images. Some are better at illustrations, product photography, text rendering, image editing, or creative styles.

But from a user’s perspective, the experience is still fragmented.

To use different models, people often need to:

• Create accounts on multiple platforms
• Learn different pricing and credit systems
• Add payment information more than once
• Understand which model is best for each task
• Learn how to write prompts
• Move generated images into other tools for editing

For experienced AI users, this may not be a major problem.

For beginners, however, the first question is often not:

What should I create?

It is:

Which model should I use?

I think that is the wrong starting point.

Most users do not care about model names at the beginning. They care about the result they want.

They want to create a social media post, a product image, a YouTube thumbnail, an illustration, or a new background for an existing photo.

That became the central idea behind Mireka:

Start with what you want to create, not with the name of an AI model.

Why I Started With Japan

Most AI image platforms are designed for English-speaking users first.

Many of them technically support Japanese prompts, but supporting Japanese input is not the same as designing a product for Japanese users.

A localized product also needs to consider:

• How features are described
• Which examples are shown
• Which image formats are commonly needed
• Which social and publishing platforms people use
• What kind of design feels familiar and trustworthy
• How pricing, credits, and instructions are presented
• Whether the Japanese copy sounds natural

For example, Japanese users may want to create images for:

• X and Instagram posts
• YouTube thumbnails
• note article covers
• LINE announcements
• E-commerce product listings
• Advertisements and promotional banners
• Anime-style or manga-style content

Simply translating an English interface into Japanese would not be enough.

Mireka is currently Japanese-first because I want to understand one market deeply before expanding into several markets superficially.

I am a solo developer from China, so building for Japan also gives me an additional challenge: I need to learn not only how to build the software, but also how to understand a different language, market, and user experience.

That challenge is one of the reasons I decided to build the product publicly.

Launching Before Everything Was Perfect

The first version of Mireka focused mainly on making the core image generation workflow work.

Users could enter a prompt, choose a model, select an image ratio, and generate an image.

Technically, the product worked.

But after reviewing the website as a new user, I noticed a major problem.

The interface explained many features, but it did not clearly answer the most important question:

What can I actually make with this?

The homepage talked too much about models and capabilities.

It expected users to understand the product before they had even tried it.

Shortly after launching, I started redesigning several parts of the website:

• The homepage structure
• The image generation interface
• The example gallery
• The product copy
• The model selection experience
• The pricing presentation
• The mobile navigation

Instead of placing AI model names at the center of the experience, I began organizing the product around user goals.

For example:

• Create a social media image
• Generate an AI illustration
• Design a product photo
• Edit an existing image
• Create an advertisement
• Make a YouTube thumbnail

This was an important lesson for me.

A feature may be technically impressive, but that does not mean users immediately understand why they need it.

What Mireka Can Do Today

Mireka currently brings several AI image generation and editing workflows into one platform.

Use Multiple AI Image Models

Users can access different AI image models through one account.

They do not need to register for a separate service every time they want to try another model.

Credits, billing, and generation history can also be managed in one place.

The goal is not to claim that one model is always the best.

Different models are useful for different tasks, so Mireka should help users choose based on what they are trying to create.

Generate Images With Japanese Prompts

Users can describe an image naturally in Japanese.

They can specify the subject, background, mood, lighting, composition, clothing, or visual style without needing to write everything in English.

This is especially important for users who are interested in AI image generation but feel uncomfortable using English prompt terminology.

Start With a Real Use Case

Instead of showing only technical aspect ratios such as 1:1, 16:9, and 9:16, Mireka connects those formats to real use cases.

For example:

• Social media post
• YouTube thumbnail
• Smartphone wallpaper
• Product image
• Article cover
• Vertical short-form content

The ratio still matters, but the purpose is easier to understand than the number.

Compare Different Models

Users can try similar prompts with different image models and compare the results.

This makes it easier to see which models are better suited for:

• Photorealistic images
• Illustrations
• Product photography
• Text inside images
• Character design
• Image editing

Model comparison is useful because marketing pages often describe every model as powerful, but the actual results can vary significantly depending on the task.

Edit Images With Natural Language

Mireka is also intended to support workflows beyond initial image generation.

Users should be able to make requests such as:

• Remove an unwanted object
• Replace the background
• Add a new element
• Change the lighting
• Adjust the colors
• Expand the image
• Modify part of the composition

My long-term goal is to reduce the need to move between multiple generation and editing tools.

Create From Examples

Writing a prompt from an empty text box can be difficult.

Mireka includes examples that users can select and modify.

An example can provide:

• A starting prompt
• A recommended image ratio
• A suitable model
• A specific use case
• A visual direction

Instead of learning prompt engineering before creating anything, users can begin with an existing example and gradually customize it.

Localization Is More Than Translation

One of the most interesting parts of building Mireka has been learning that multilingual development is not mainly a translation problem.

Translation is only one layer.

A real multilingual product also needs localization at the level of intent.

For example, a landing page created for English-speaking users may focus on concepts such as personal branding, LinkedIn content, or Etsy product images.

A Japanese landing page may need different examples, terminology, visual references, and content platforms.

Even seemingly simple UI labels can become difficult.

A word may be grammatically correct but still feel unnatural in a SaaS interface.

Pricing pages are another example.

The best way to explain monthly credits, annual billing, one-time credit packs, and generation limits may differ between markets.

This means I cannot simply build the Japanese version and later run every string through a translation system.

For each language, I will need to reconsider:

• Search intent
• Landing page structure
• Example prompts
• Product use cases
• SEO keywords
• Interface terminology
• Pricing presentation
• User expectations

The Current Technology Stack

Mireka is built with a modern TypeScript-based stack.

The main technologies include:

• Next.js
• TypeScript
• Tailwind CSS
• Supabase
• Drizzle ORM
• Stripe
• Cloudflare
• Multiple external AI model APIs

Using external AI providers makes it possible to support different models, but it also introduces additional engineering challenges.

Each provider may have different:

• Request formats
• Image size rules
• Processing times
• Credit costs
• Error responses
• Moderation rules
• Output formats
• Callback systems

A large part of the work is not simply calling an image generation API.

It is building a consistent user experience on top of several inconsistent systems.

The user should not need to understand how every provider works internally.

They should only need to choose what they want to create.

Building in Public

I am still deciding what to share as part of the build-in-public process.

Some topics I plan to write about include:

• How I choose which AI models to integrate
• How I design a multi-model credit system
• How I approach Japanese SEO
• Mistakes in the first homepage design
• How I organize prompt examples
• How I localize product copy
• What users actually generate
• Which features people ignore
• Pricing experiments
• Expanding from Japanese to English

I also want to share failures, not only successful launches and revenue screenshots.

A large percentage of product development is changing things that looked correct when they were first built.

What Comes Next

For now, my main priority is improving the Japanese version.

I want to make the first-time experience clearer, add more practical use cases, improve the example gallery, and make model selection easier to understand.

After that, I plan to launch an English version.

Other languages may follow, but I do not want to release many poorly localized versions at the same time.

The multilingual plan will probably look like this:

1. Improve the Japanese product experience
2. Identify workflows that work across markets
3. Build the English version
4. Adapt examples and landing pages for English-speaking users
5. Expand into additional languages gradually

The English version will not simply be a translated copy of the Japanese website.

It should reflect the platforms, search behavior, creative needs, and terminology of English-speaking users.

I Would Appreciate Your Feedback

Mireka is still an early-stage product, so feedback is especially valuable right now.

I would love to know:

• Is the product’s purpose clear when you first open the website?
• Does the image generation workflow feel easy to understand?
• Is choosing between different AI models confusing?
• Which image generation or editing features would you expect?
• Which models would you like to see supported?
• What would make you choose a multi-model platform instead of using each provider separately?
• What parts of the website feel unclear or unnecessary?

You can try the current Japanese version here:

Mireka.jp

I will continue sharing the development process, design decisions, mistakes, and lessons as the product evolves.

Thanks for reading!