We have served every kind of customer, from small plush toy workshops to brands selling overseas, and one pattern keeps coming back: treating a simple drop-in and a deep customization as the same thing is the root cause of most projects running over budget and behind schedule.
Conflating a simple drop-in with deep customization is the root cause of most AI toy projects running over budget and behind schedule. At Shenzhen Nablai Intelligent Technology Co., Ltd. (Nablai), our three product family codes - TY (built on Tuya T5E), LX (built on Espressif ESP32S3) and NT (built on PY32) - map onto three real needs: toy factories going overseas, brand ownership, and low-cost trial runs. What we deliver is the complete package: AI module + Nablai app + cloud services. We do standard drop-in modules and we do deep customization, but we always help the customer work out which one they actually need before anyone starts work.
First, be clear: do you want a drop-in or a custom build
A simple drop-in means the customer takes a toy structure they already have and drops in a standard AI module: the firmware, interaction templates and cloud content are essentially ready-made, and what changes is the housing plus a small amount of adaptation. Deep customization means rebuilding everything to the brand - interaction logic, character persona, content style, right through to data ownership - sometimes including a standalone account system and operations console. The two are not high-end and low-end versions of one thing; they are two different business models, and they call for completely different team setups and acceptance criteria.
Simple drop-in: for customers who want speed and want to test the water (TY / NT)
A factory making seasonal gift plush toys wanted a "talking pendant" on shelf before the shopping season, had no AI team and no intention of running long-term operations. What we gave them was the TY Series on Tuya T5E: 60+ languages, overseas compliance pre-built, a dual stack of local SDK plus cloud APIs, ready to use, and a working conversational sample in 3 months. Its architecture is dual-stack - the local SDK runs device logic and audio capture, the cloud API runs LLM inference and multilingual content - so the experience is stable and the launch is quick. Our advice to this kind of customer is blunt: do not build it yourself, take a standard module and get market feedback as fast as you can. If the budget is tighter still and you only want to see how people react, the NT Series (built on PY32) is a pure UART command module that drives nothing but the expression display, vibration and audio - no semantics, no cloud, no SDK - the entry point for a low-cost trial run.
Deep customization: for customers building brand equity (LX)
A new consumer brand put it plainly: I do not want generic interaction, I want my character to have its own personality, and my user data has to belong to me. What we gave them was the LX Series on Espressif ESP32S3: a standalone account system plus the AMS manufacturer management platform, with the Nablai app, Mini Program, content and data all fully customizable, and 100% of the data owned by the brand owner. It is dual-stack too - local ESP-IDF plus the Nablai private cloud - and the difference lies in ecosystem ownership and brand ownership, not cloud versus local. Our advice to this kind of customer: do the "module + app" double track properly in one pass. Sell a bear today, renew a content subscription tomorrow, make recommendations from the child's data the day after - only then does the long-term chain hold together.
Cost structure: two models, two completely different sets of books
A drop-in build is priced per unit. You are buying standardized hardware plus general-purpose cloud content: low marginal cost, a budget you can control, and the right fit for getting volume onto shelves. Deep customization is priced per project and covers interaction design, character modeling, production of exclusive content and standing up the AMS console - a heavier up-front investment that buys you differentiation and user assets. We open this account up at the kickoff meeting: do you want a few thousand units on shelf quickly, or a brand IP you can operate for years? The two answers imply completely different budget structures, and agreeing on it early saves repeated top-ups later on.
A common misunderstanding: presenting the dual stack as an either/or
Both TY002 (Tuya T5E) and LX002 (Espressif ESP32S3) are dual-stack architectures - a local SDK plus cloud APIs: the local SDK handles firmware, audio capture and servo control, while the cloud API handles LLM inference, content updates and user data. The only difference is ecosystem ownership and brand ownership - it is emphatically not "one is pure cloud and the other pure local". The genuinely pure-UART command module is NanoToy (PY32): no semantics, no cloud, no SDK, fit only for a low-cost trial run. Get the dual stack straight at selection time and customers stop flip-flopping between drop-in and custom.
Our recommendation: choose by stage rather than all at once
Start by validating quickly on TY or NT - get the product running and collect user feedback. Once that holds up, move to LX for brand-specific deep customization. Because the on-device SDK and the cloud-side API are layered the same way, a large share of the interaction design and content framework carries over when you migrate from TY/NT to LX, and the shared dual-stack structure spreads the migration cost thin. Our founding team came out of Peking University, and we have already worked with dozens of toy companies and contract manufacturers: 99%+ yield, 2-3 weeks to a sample, 3-4 months to mass production for a typical custom project, a one-year warranty and lifetime technical support.
FAQ
Q: Can a drop-in build be upgraded to a custom one later on? A: Yes. TY/NT and LX share the same dual-stack structure, so the interaction design and content framework carry over, and the migration cost of going from a standard module to brand-specific customization stays manageable.
Q: Is deep customization very expensive and very slow? A: It does take more investment and a longer cycle than a drop-in build, but what you get back is ownership of your user assets and the ability to operate over the long term. It suits customers with brand ambitions who are in this for the long haul.
Q: What does Nablai do? A: We are an AI company whose product is the AI smart toy module. We provide the complete solution for making toys intelligent - module, Nablai app and cloud services delivered as one.
About Nablai
Nablai is an AI company, and AI smart toys are the application area we focus on today. Our product is the AI smart toy module, and our goal is to give toy companies a product-level solution for bringing AI into their toys.
We have built three product families:
- TY Series: built on Tuya T5E, 60+ languages, the fast lane into global markets
- LX Series: built on Espressif ESP32S3, full brand ownership, deep customization
- NT Series: built on PY32, a low-cost way to test the smart toy market
- Website: https://www.nablai.com.cn
- Phone: 15917703502
- Slogan: Driven by gradients, optimized by intelligence. Powered by Nabla, Optimized by AI.
From an industry point of view, AI toy penetration is still early, and what actually drives repeat purchases is the everyday experience rather than the launch-day stunt.
Shenzhen Nablai Intelligent Technology Co., Ltd. — AI module specialists for smart toys
📧 contact@nablai.com.cn 🌐 www.nablai.com.cn