What If Every Story Could Be Seen?
The Evolution of EdTech — Part 12 of 13
I spent eleven chapters tracing a hundred years of educational technology — from Pressey's mechanical teaching machine to Skinner's programmed instruction, from Sesame Street to Khan Academy, from the MOOC tsunami to the AI revolution. Through all of it, one pattern repeated with dispiriting consistency: the technology improved, the promises expanded, and the reading crisis deepened.
In 2024, I decided to stop documenting the problem and start building a solution.
This is not the part where I claim to have solved the crisis. Nobody has. The reading crisis is a systemic failure a century in the making, and no single product will fix it. But I believe — with the conviction of someone who spent years studying both the technology and the problem it should address — that we finally have the tools to build something that has never existed before: a reading platform that bridges the gap between text and visual comprehension, using AI that runs locally, privately, and in real time.
I called it Readify. It's live at app.ireadifybooks.com. And it works.
The Core Insight
The insight behind Readify isn't complicated. It follows directly from the research I discussed in the previous chapter: reading comprehension is fundamentally a visual process.
When proficient readers engage with a novel, they're not just processing words. They're constructing a film in their minds. They see the forest. They feel the weather. They picture the expressions on characters' faces. This mental imagery — what researchers call a "situation model" — is what transforms reading from a decoding exercise into an experience. It's what makes books immersive, memorable, and meaningful.
Struggling readers often lack this visualization capacity. They process text at the surface level — word by word, sentence by sentence — without constructing the deeper mental representation that makes reading comprehensible and engaging. Reading, for them, is like listening to a song in a language they barely speak. They hear the sounds without feeling the meaning.
The question I asked was straightforward: what if AI could build the mental image for the reader?
Not as a replacement for imagination — but as scaffolding. A visual companion that renders what the text describes, making the abstract concrete and the unfamiliar recognizable. A tool that meets struggling readers where they are and helps them develop the comprehension skills to eventually read without it.
How It Works
Readify is an AI-powered reading platform that transforms any EPUB book into a visually enriched reading experience.
Text analysis via local language model. I use Qwen3-8B, an open-source large language model from Alibaba, running locally via vLLM on a dedicated GPU. The model reads each section of text and generates a rich visual description — a prompt capturing the setting, mood, key elements, and artistic style of the scene.
Image generation via local diffusion model. I use SDXL (Stable Diffusion XL), an open-source image generation model, running locally on a second dedicated GPU. The system produces illustrations at native 1024x1024 resolution that depict the content of each text section.
Visual rhythm alongside text. Every twenty blocks of text — roughly a page or two — a new illustration appears, creating a visual cadence that guides the reader through the narrative. The illustrations aren't generic stock imagery. They're generated specifically for the content being read, in the reader's chosen style — oil painting, anime, photorealistic, children's book illustration, and more.
Complete local operation. No data leaves the device. No text gets transmitted to a cloud server. No reading habits are tracked by a third party. Both models run on local GPUs, ensuring complete privacy — which matters enormously when your users include children, adult learners, and anyone who'd rather their reading habits not be monetized.
The Technology Stack
I built Readify on a dual-GPU workstation with two NVIDIA RTX 4090s — one dedicated to language processing, one to image generation:
- GPU 0: vLLM + Qwen3-8B handles all text analysis, prompt generation, and text transformations (simplification, vocabulary assistance, summarization).
- GPU 1: OmniA v1.0 + SDXL Refiner generates illustrations. The base model handles 80% of the denoising; the refiner adds fine details in the final 20%. Total generation time: about 4.6 seconds per image.
- Deterministic caching: Images are cached based on a hash of book content, style, and parameters. Same inputs produce the same images every time, for every user. Once generated, images serve instantly.
- Speed optimizations: VAE tiling, channels-last memory format, xFormers attention, and token merging combine to maximize throughput without sacrificing quality.
The result: contextually relevant, high-quality illustrations in under five seconds — fast enough to feel natural, detailed enough to actually help with comprehension.
What Makes This Different
I want to be precise about what Readify is and isn't, because the EdTech industry has a credibility problem and I have no interest in adding to it.
Readify is not an AI tutor. It doesn't quiz you. It doesn't track progress on a dashboard. It doesn't claim equivalence with a human reading specialist. Khanmigo handles tutoring well. MagicSchool handles teacher tools well. Readify does something different.
Readify is a reading experience platform. It transforms the act of reading — making it more visual, more immersive, more engaging. The hypothesis: readers who can see what they're reading will understand more, retain more, and read more. This is grounded in dual coding theory, multimedia learning research, and the basic observation that illustrated books have been helping struggling readers for centuries. What's new is that AI eliminates the bottleneck. Previously, illustration required human artists — expensive, slow, limited to a few titles. Readify can illustrate any book, in any style, in real time.
Readify includes text transformations. The language model can simplify vocabulary, rewrite passages at lower reading levels, summarize sections, and define unfamiliar words — all in context, all on demand. A reader who hits a difficult passage can access a simplified version without leaving the page. The text meets the reader; the reader doesn't need to meet the text.
The Capital-Efficient Model
One thing that separates Readify from most EdTech startups is how I think about cost.
Most AI-powered educational products rely on cloud-based models — primarily OpenAI's GPT-4 or equivalent — charging per API call. Every interaction costs money. Those economics work for companies serving affluent users at premium price points. They don't work for reaching the 130 million Americans who read below grade level and who are, by definition, disproportionately low-income.
Readify runs entirely on local, open-source models. Once I've purchased the hardware, the marginal cost of each interaction is essentially the cost of electricity. A single dual-GPU workstation can serve hundreds of concurrent readers. At scale, cost per reader is a fraction of what cloud-dependent competitors pay.
This matters because the people who need reading technology most are those who can least afford it. A $20-per-month subscription will never reach the population that needs it most. A system built on open-source models and commodity hardware has a path to affordability that cloud-dependent approaches simply don't.
The Mission
I want to be direct about motivation, because in a field crowded with venture-backed startups optimizing for growth metrics, it matters.
Readify is education-first. The goal isn't to build the next unicorn. The goal is to build technology that helps people read — people who have been failed by the educational system, overlooked by the EdTech industry, and left behind by the AI revolution.
The motto is: "Every story deserves to be seen."
It's a statement about the product — visualizing text — and about the mission — making reading accessible to everyone. The reading crisis won't be solved by better phonics apps or more standardized tests. It'll be solved, if it's solved, by making reading itself a richer, more engaging, more comprehensible experience.
What the Research Suggests
I want to handle claims carefully here. Readify hasn't been tested in a randomized controlled trial. The evidence for its specific approach — AI-generated illustrations as a reading comprehension tool — is theoretical, not yet empirical. The product is new. The research will take time.
But the theoretical foundations are strong. Richard Mayer's principles of multimedia learning — that people learn better from words and pictures together than from words alone — are among the most replicated findings in educational psychology. Allan Paivio's dual coding theory has been validated across hundreds of studies over five decades.
What I've built is the practical implementation: using generative AI to create the visual component on demand, for any text, at any scale. The theory predicts this should help. The technology makes it feasible. The empirical validation will follow.
In the meantime, the demo is live. Try it. Upload a book you love. Watch the AI turn words into images. And consider whether this is the kind of tool that could help a struggling reader stay engaged long enough to build the skills they need.
I think it is. But thinking isn't evidence. That's why the demo is public and why the research agenda is a priority.
Where This Fits
Readify isn't a competitor to Khan Academy, Khanmigo, MagicSchool, or Duolingo. It occupies a space those platforms don't: the intersection of reading, visual comprehension, and AI-generated content.
I have genuine respect for what Sal Khan has built. Khan Academy's commitment to free education is one of the great achievements of the EdTech era. Khanmigo's approach to AI tutoring is thoughtful and evidence-driven. MagicSchool's tools for teachers are practical and sorely needed.
Readify does something different. It transforms the reading experience itself. It's a tool for readers, not test-takers. For stories, not standards. For comprehension, not compliance.
And it's built on the conviction that the reading crisis won't be solved by teaching people to read harder. It'll be solved by making reading better — richer, more visual, more engaging, more human.
Look. The technology is finally good enough. The need has never been greater. The question is whether we direct the most powerful tools in human history toward the most fundamental problem in education — or whether, like every previous wave of EdTech, the tools flow toward those who need them least.
Sources
- Mayer, R.E. "Multimedia Learning." Cambridge University Press, 2009.
- Paivio, A. "Mental Representations: A Dual Coding Approach." Oxford University Press, 1986.
- "Reading the Numbers: 130 Million American Adults Have Low Literacy Skills." APM Research Lab. https://www.apmresearchlab.org/10x-adult-literacy
- "2024-2025 Literacy Statistics." National Literacy Institute. https://www.thenationalliteracyinstitute.com/2024-2025-literacy-statistics
- Van den Broek, P. "Cognitive Processes in Comprehension of Science Texts." Science Education, 2010.
- "Explore Results for the 2024 NAEP Reading Assessment." Nation's Report Card. https://www.nationsreportcard.gov/reports/reading/2024/g4_8/
What could AI do for your classroom?
Readify uses local AI to adapt reading material to each student's level — generating illustrations, simplifying vocabulary, and tracking comprehension in real time. No student data leaves the building.
Liked this article?
Get the full Evolution of EdTech series delivered to your inbox, plus weekly insights on AI in education.
Subscribe to Readify Branches