We've been hard at work behind the scenes at dokkei, and we're excited to share a major development update: the core of the dokkei image generation engine is now working reliably.
That means we're officially past the "can this work?" stage. Now we're focused on refining the system so that every illustrated book feels consistent, immersive, and natural to read.
The core image generation API is live
At the heart of dokkei is an API that reads a book and generates illustrations for key scenes as you read.
Over the past few weeks we've made major progress on this system. The core pipeline is now able to:
- parse EPUB books
- identify important scenes in the story
- generate illustrations that match those scenes
- maintain visual consistency across images
Most importantly, the system is now generating images reliably and consistently. When a book goes through the pipeline, the results are stable and predictable — which is exactly what we need before scaling the platform.
This milestone gives us a strong technical foundation to build on.
What we're improving next
With the core system in place, we're now focusing on refining two areas that will make the experience feel truly seamless.
Character consistency
One of the hardest problems in AI illustration is ensuring that characters stay recognisable across multiple scenes.
If a character appears early in a book, they should still look like the same person dozens of scenes later. We're currently fine-tuning how dokkei tracks and reinforces character traits so that:
- characters keep the same face and visual features
- clothing and defining traits remain consistent
- the overall illustration style stays cohesive throughout the book
This is essential to making the images feel like they belong to a single illustrated story.
Smarter scene distribution
We're also refining how scenes are selected and spaced throughout a book.
Not every paragraph needs an illustration — but the right moments absolutely should have one. We're improving how dokkei decides:
- which moments are visually meaningful
- where illustrations should appear within the reading flow
- how frequently images should appear to enhance the story without overwhelming it
The goal is simple: illustrations should feel natural, not forced.
Building the dokkei e-reader app
Alongside the API development, we've also started building the dokkei e-reader app for iOS and Android.
Rather than relying only on a web reader, we're creating a native mobile experience designed specifically for illustrated books.
This allows us to deliver:
- smooth page transitions
- fast image loading
- offline reading
- a reading interface designed for immersive storytelling
Our goal is to make dokkei feel like the best e-reader you've ever used — with the added magic of AI illustrations.
Why native apps matter
Native apps allow us to integrate illustrations directly into the reading experience.
Instead of images feeling like attachments or interruptions, they become a natural part of the story flow. As you read, illustrations can appear exactly when the scene unfolds.
That's the experience we're building toward.
What comes next
Over the coming weeks we'll be focused on:
- improving character tracking and visual consistency
- refining scene detection and illustration timing
- continuing development of the iOS and Android reader apps
- polishing the full reading experience from upload to illustrated book
Every improvement brings us closer to making AI-illustrated reading feel seamless and magical.