Student's $83K AI Video Site: Real Innovation or Open-Source Wrapper?

A news report claimed a sophomore student built an AI video generation website from scratch, attracting 20,000 users and generating 600,000 RMB (approx. $83,000) in two months. The author, a developer known as Xiao Meng, analyzes this claim with skepticism.

Technical Architecture: Relay Station vs. Model Training

The phrase "hand-rolled AI" suggests training a model from zero, but AI products can follow different technical paths: training on open-source models, calling third-party APIs, or acting as a relay station that forwards requests to upstream models and manages billing, queuing, and support. The original report mentioned "using open-source models for training" but provided no verifiable technical details. Existing evidence neither proves the student built a model from scratch nor confirms the site is purely a relay station.

Cost Structure and Student Discounts

The author notes a client's AI video site incurs significant costs — official APIs cost about 1 RMB per minute. The student's claimed cost reduction to "a few percent of industry rates" raises questions. Student discounts exist (e.g., Google's one-year free AI package for eligible students), but such benefits are limited by eligibility, region, and terms. If a platform relies on discounted accounts, trial credits, or temporary subsidies, its price advantage may vanish when promotions end. Moreover, membership rights, API quotas, commercial-use permissions, and resale rights must be verified separately; "students can use for free" does not equal "can be resold commercially." Allegations of organizing mass student registrations to harvest benefits remain unproven and should be treated as leads for investigation, not conclusions.

Open-Source Project Comparison: MagicalCanvas

The author observed striking similarities between the student's site and the open-source project MagicalCanvas (GitHub:

https://github.com/abutao/MagicalCanvas

). Logs and the node-based canvas interface appear identical. MagicalCanvas combines AI content generation with video editing, offering a node canvas, batch generation, multi-track editing, and configuration of external model endpoints, keys, and names. The student's site displays many models; optimizing that many models individually seems implausible.

The author's own AI video site screenshot is also included for reference.

Product Value: Integration vs. Model Access

Does a product that integrates others' models still have value? Yes, if it saves creators substantial work: script breakdown, prompt writing, asset generation, file organization, and editing. A tool that stitches these steps together, reduces repetitive operations, and stabilizes output provides measurable service value. This value must be demonstrated through features and results, not marketing claims.

Need for Transparent Metrics

The 600k revenue figure requires breakdown: over what period? How many paying users? What are model and server costs? Why do users keep paying? These questions are more instructive for aspiring AI product builders than the headline "student earns 600k with AI."

We can applaud a young person's initiative to identify demand, build a product, and serve users, while continuing to scrutinize technical origins, cost structure, and revenue accounting. The hope is that students create genuinely useful tools, not just hype.