She Built a Manga App in 24 Hours Without Writing Code

Tina Huang's drawing ability, by her own admission, rates somewhere around negative three on a ten-point scale. Yet in seventeen hours and thirty-two minutes, she produced a complete manga called "The In Between," using a custom application she built from scratch without writing a single line of code.

This isn't another breathless AI demonstration where someone types a prompt and declares victory. Huang documented the actual process—the failures, the workarounds, the hour spent wrestling with dialogue bubble rendering, the inexplicable difficulty of getting AI to draw a simple ballpoint pen without two clips. What emerges is a clearer picture of where these tools actually land between hype and utility.

The Mechanics of No-Code Development

Huang used Bolt, a platform that generates functional applications from natural language descriptions. She described what she wanted—character creator, panel generator, page composer—and the system produced working code. When features broke or behaved unexpectedly, she described the problems in plain English and the system adjusted.

The first attempt at character generation failed immediately. The front and back views of a character looked like different people entirely. "This is like mostly a prompting thing, though," Huang noted, adjusting the system prompt rather than the underlying code. The second attempt worked.

This pattern repeated throughout: request feature, test, describe problem, iterate. Dimensions rendered incorrectly in the page composer. Text generation mixed Japanese and English unpredictably. Each issue got addressed through description rather than debugging. The technical barrier to entry had shifted from "can you code" to "can you articulate what's wrong."

What Actually Works

The completed manga studio offered three core functions. Users could generate character sheets with consistent front and back views. They could create individual manga panels either from text descriptions or by uploading rough sketches—Huang's crude stick figure of a person holding paper became a properly rendered manga panel. The page composer assembled multiple panels into full pages.

Huang also added what she called "lazy mode"—full page generation from pure text prompts. No sketching, no panel-by-panel composition. Just describe the scene and the system renders an entire page. For her manga's opening sequence—Tokyo office building at dusk, protagonist Ayaka refilling her coffee, sudden transportation to another world—this approach produced usable results in minutes.

"Oh my gosh, if this works, that would be very interesting," she said before generating her first full page from a text prompt. Then: "Oh, hey, not bad. Can we just give the AI some credit here?"

Where It Still Breaks

Text rendering nearly derailed the entire project. The AI could generate speech bubbles or overlay text, but matching the quality and style of the manga panels proved difficult. After an hour of attempts to solve this programmatically, Huang abandoned the technical solution and used Canva instead. Sometimes the lowest-tech answer is the right one.

Character consistency across panels required careful management. Multiple characters in a single scene needed extra attention. The system excelled at single-character panels with clear reference images but struggled when the complexity increased.

And that ballpoint pen. Simple objects without human features confounded the character sheet generator, which expected people. Huang eventually switched to Google's Gemini for that specific asset. Even streamlined workflows hit friction points that require human judgment about which tool fits which task.

The Broader Pattern

Before building anything, Huang took an entire course on manga creation. This matters more than the technology choices. The tools enabled execution, but understanding manga's visual language, panel flow, and narrative structure came from traditional learning. AI didn't replace expertise—it made expertise accessible to someone without drawing ability.

Her story concept drew from Lewis Carroll's "Through the Looking-Glass," filtered through isekai manga conventions. She wanted a perfectionist protagonist confronting a world where logic fails and efficiency backfires. The AI generated character designs and rendered scenes, but the narrative architecture came from Huang's synthesis of literary and manga traditions.

This distinction runs throughout the project. The AI never suggested that generating a simple ballpoint pen would be harder than rendering complex characters. It couldn't tell her when to abandon technical solutions for Canva. It didn't know that character consistency degrades in multi-character scenes. That knowledge came from testing, from prior experience with similar tools, from the accumulated judgment that no prompt can encode.

What This Means for Creative Work

Huang positioned this as workflow improvement rather than revolution. "This wider trend of being able to improve your workflows by building your own apps," she said, noting the potential to sell successful tools as SaaS products.

That framing matters. This isn't about AI replacing illustrators or developers. It's about whether people with domain knowledge but lacking specific technical skills can now build tools that serve their needs. Huang knew what manga required. She couldn't draw it or code it. The gap between knowledge and execution has narrowed, though not disappeared.

The seventeen hours and thirty-two minutes tell their own story. Not instant. Not trivial. But achievable within a day for someone willing to learn the tools, iterate through failures, and know when to switch approaches. The time investment was in problem-solving and iteration, not syntax and debugging.

Whether this democratizes creative tooling or simply adds another layer of technical gatekeeping depends partly on how these platforms evolve and partly on whether users approach them with realistic expectations. Huang's documentation of actual friction points—the parts that didn't work, the workarounds required, the hour lost to text rendering—provides more useful information than a dozen polished demos.

The manga exists. You can read it. That's the test that matters.

—Bob Reynolds