DEV Community: jacksam

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

jacksam — Thu, 23 Jul 2026 06:41:00 +0000

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:

Improve the Japanese product experience
Identify workflows that work across markets
Build the English version
Adapt examples and landing pages for English-speaking users
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!

How I Turned a Single LinkedIn Headshot Into a 6-Second Personal Branding Intro Video With AI

jacksam — Tue, 16 Jun 2026 04:02:58 +0000

Most people update their LinkedIn profile photo and stop there.

I did the same for years.

Then I started wondering:

What if the same photo could become a short personal branding video?

Not a talking avatar.

Not a fake AI spokesperson.

Just a subtle motion clip that feels like a real camera shot.

So I tried an experiment.

I took a single LinkedIn headshot and turned it into a 6-second intro video using an AI photo-to-video workflow.

Here is exactly what I learned.

The Starting Point

I used a simple professional headshot:

Neutral background
Looking slightly off camera
Good lighting
High resolution

The image was originally created for LinkedIn, but it also appeared on:

personal website
conference speaker page
GitHub profile
newsletter author profile

Instead of creating a new video from scratch, I wanted to reuse the image I already had.

Why Most AI Animations Look Weird

The first few attempts looked terrible.

My prompts were things like:

make this person move naturally

The results were:

exaggerated head movement
unnatural facial expressions
strange eye motion
distracting camera movement

The problem wasn't the model.

The problem was the instruction.

I realized that good photo-to-video generation is mostly about motion design.

The photo already contains the composition.

The AI only needs to create believable movement.

The Prompt Structure That Worked

Instead of describing the person, I started describing the shot.

My final prompt looked something like:

Subtle natural blinking, slight breathing motion, gentle movement in hair, slow cinematic camera push in, professional lighting preserved, realistic facial details, no exaggerated expression changes, natural motion only.

This immediately produced better results.

The biggest improvement came from limiting movement instead of adding movement.

The Three Types of Motion

I found that every successful result contained three motion layers.

1. Subject Motion

Very small movement.

Examples:

blinking
breathing
slight posture adjustment

Anything larger started to look artificial.

2. Environmental Motion

Tiny background activity creates realism.

Examples:

moving light
drifting particles
subtle depth changes

Even minimal environmental motion makes a static image feel alive.

3. Camera Motion

This was the most important layer.

A slow push-in instantly made the clip feel like real footage.

Without camera movement, the animation still felt like a photo.

The Final Result

The finished video was only six seconds long.

Nothing dramatic happened.

The person didn't talk.

There was no voiceover.

The camera slowly moved forward.

The subject blinked once.

The lighting shifted slightly.

That was enough.

The result felt more like a cinematic introduction than an AI animation.

Where This Works Surprisingly Well

I now use short photo-to-video clips in several places:

LinkedIn profile content
speaker introductions
portfolio websites
conference landing pages
personal brand reels
newsletter promotions

Video naturally attracts more attention than static images, which is one reason personal branding experts often recommend incorporating video into online profiles and content.

What I Would Do Differently

If I repeated the experiment, I would spend more time preparing the original image.

The quality of the source photo matters more than most people think.

A strong composition produces significantly better motion results.

That's also why many experienced AI creators focus on building a good image before trying to animate it.

A Simple Workflow Anyone Can Follow

My workflow is now:

Create or choose a strong photo.
Define subject motion.
Define environmental motion.
Define camera motion.
Add preservation instructions.
Generate multiple versions.
Keep the most natural result.

If you're new to AI video creation, I found this guide on creating intentional photo-to-video prompts particularly useful:

Photo to Video AI Workflow:
https://phototovideoai.co/blog/how-to-turn-photos-into-ai-videos

It explains how to think about motion before generating the video.

For generating the actual animation, I used:
https://phototovideoai.co/

It allows you to turn a still image into a short AI-generated video using motion prompts and camera movement instructions.

Final Thoughts

The interesting part wasn't the technology.

It was realizing that a single photo can become a reusable video asset.

Most of us already have profile photos sitting in folders, on LinkedIn, or on our websites.

Adding subtle motion can make those images feel much more alive without requiring a camera, microphone, or video editing workflow.

Sometimes a six-second clip is all you need.

I Built a Multi-Model AI Image & Video Platform — Here's What I Learned

jacksam — Sun, 03 May 2026 13:01:49 +0000

A few months ago I started building Bananai — a platform that
lets you generate images, edit photos, and create AI videos all in one place without
switching between five different tools and five different billing dashboards.

Here's what I actually learned integrating 10+ models end-to-end as an indie dev.

Why I built it

My problem was embarrassingly mundane: I needed to test GPT Image 2 against Nano Banana
Pro for an e-commerce client's product shots. I had four browser tabs open, four
different logins, four different credit top-ups, and I was copy-pasting the same prompt
over and over.

The obvious solution was to wrap them in a unified UI. What I thought would take a
weekend ended up being a real product — because model integration is the easy part.
Everything else is the hard part.

1. Picking the right model for the task is not obvious

The first mistake I made was treating "best model = best output for every use case." That's
wrong. Here's how I actually think about it now:

Task	Model choice	Why
Fast social content iteration	Nano Banana 2	2–5s per image, good enough quality
Client deliverables / 4K output	Nano Banana Pro	Max quality, character consistency
Text in images (logos, UI mocks)	GPT Image 2	Best text rendering accuracy by far
Animate a still product photo	Seedance / Veo	Different motion styles, test both
Background removal + style transfer	Nano Banana (editing mode)	Natural language edit instructions

The practical takeaway: expose model selection to the user early. Users figure out
their own preferences faster than any default you set. I wasted a month pre-selecting
defaults before realizing users just want the picker.

2. The cost structure is wilder than you'd expect

Before building this I assumed model pricing was roughly proportional to quality. It's not.

GPT Image 2 — ~$0.006/image at standard quality. Shockingly cheap for what it outputs.
Nano Banana 2 — fast and economical, great for high-volume generation.
Video models — anywhere from 10x to 50x more expensive than image per output second. Budget for this separately. Video is a different product category economically.

The thing that trips up most builders: you need to model your credit burn rate for
different user behavior patterns, not averages. A power user generating 4K video clips
will burn through credits 30x faster than someone doing quick image edits. If you price
on averages, power users kill your margin.

My approach: track generation cost per user session, flag sessions above the 95th
percentile, and cap or up-sell those users. Works better than blanket rate limits.

3. Async generation UX is harder than sync

Most image models are fast enough to feel synchronous (~3–5 seconds). But video generation
can take 20–60 seconds depending on model and duration. That's a different UX contract.

What didn't work:

Spinner with no feedback → users thought it crashed and refreshed
Showing "estimated time" → estimates were wrong enough to frustrate people more than no estimate

What actually worked:

Progress bar that moves in two phases: "queued → processing" with distinct visual states
Showing a low-res preview/thumbnail as soon as it's available while full resolution renders
Email/notification when a long video finishes (reduces page-staring behavior)

For image generation, the UX expectation is basically instant. Anything over 8 seconds
feels broken to users, even if technically the model is just slow. Cache aggressively,
pre-warm where possible.

4. Prompt UX is underrated

Most AI image tools show you a blank text box. That's terrible for conversion and
terrible for retention — new users don't know what to type and leave.

What I added instead:

const promptSuggestions = [
  "🎨 Product Hero Shot",
  "🖼️ Remove Background",
  "🎬 Cinematic Scene",
  "✨ Style Transfer"
];

These aren't just UI sugar. Each one loads a pre-filled prompt template with the
right model pre-selected and the right settings pre-configured. Click "Product Hero
Shot" and you get an image-to-image flow with the right aspect ratio for e-commerce, not
a blank canvas.

Conversion from landing → first generation went up significantly after this. Users who
generate at least one image on their first visit retain at 3x the rate of those who
don't.

5. Multi-model output comparison is the killer feature nobody talks about

The most popular session pattern I see in analytics: user generates with one model, then
immediately tries the same prompt with another model to compare. This is the workflow
professional designers actually use — they're not loyal to a model, they want to see options.

Building side-by-side comparison mode is on the roadmap. If you're building a similar
tool: this is worth prioritizing early. It's also the stickiest feature because it
locks users into your platform (they can't compare across separate tools with one click).

What's live now

Bananai currently has:

Image generation: Nano Banana 2 & Pro, GPT Image 2, Grok Imagine, Midjourney, Seedream, Wan 2.7
Image editing: background removal, style transfer, inpainting, upscaling — all via natural language
Video generation: Veo 3.1, Seedance 2.0, Wan 2.7 Video, Grok Imagine Video
Free credits on sign-up, daily check-in credits, no credit card required to start
GPT Image 2 at $0.006/image — cheapest I've found anywhere

If you're building something with AI image generation or just want to test models
without juggling multiple accounts, give it a try.

What I'd do differently

Start with one model, nail the UX, then add models. I added too many too fast and spread the QA effort too thin.
Instrument cost tracking from day one. I retrofitted it and lost two weeks of data.
Don't underestimate video. It looks like "image but animated" but it's actually a completely different infrastructure, moderation, and pricing problem.

Happy to answer questions about any of this — model integration, credit system design,
or the UX decisions. Drop them in the comments.

Built with Next.js, deployed on Vercel, with Cloudflare for CDN/image resizing.
Backend is a monolith I'm slowly ashamed of.