AI Lighting Scene SolutionOn-App AI
### Pain Point Analysis
In smart lighting scene applications, users (including general consumers, designers, and sales/installation teams) want to quickly create professional, aesthetically pleasing lighting atmospheres adapted to their spaces. However, the current lighting scene creation process presents significant obstacles:
- Cumbersome Operations: Users must complete multiple steps sequentially, including room selection, device filtering, and lamp-by-lamp parameter adjustment, making the process lengthy.
- High Professional Threshold: Users need basic knowledge of color temperature, brightness, dynamic effects, etc., otherwise it's difficult to achieve ideal results.
- Low Efficiency: When facing multi-device batch configuration, manual adjustment is extremely time-consuming, failing to meet rapid delivery or high-frequency creation needs.
- Experience Dependency: Without intelligent recommendations, users can only try by feeling, resulting in high trial-and-error costs.
### Solution
On-App AI Lighting Scenes — Using Tuya's self-developed lightweight spatial lighting scene generation model, professional lighting solutions are generated with one tap locally on the phone (on-device):
- Fully local inference, no internet required, supports offline use
- Millisecond-level response (average < 10ms), achieving "instant fingertip results"
- Zero cloud inference costs, drastically reducing operational expenses
- Batch generation of multiple professional lighting solutions, covering various scene styles such as reception, movie watching, sleeping, party, etc.
> 💡 **Try It Now**: Open **Smart Life App** / **Tuya App** (v7.1.0 or above), scan the QR code below to preview:
### Technical Advantages
✔ **Efficient Inference**
- Compatible with mainstream iOS/Android devices:
- Based on TensorFlow Lite deployment, compatible with mainstream iOS/Android devices:
- High-end devices: 5~8ms
- Mid-range devices: 8~12ms
- Low-end devices: 10~15ms (still much faster than human visual perception threshold)
✔ **Low Cost**
- All inference completed on-device, no token consumption, no GPU server load
- Supports large-scale concurrent user access, marginal cost approaches zero
✔ **More Secure**
- **Fully Local Processing**: Lighting parameter generation, core data processing, and calculations are all closed-loop within the device
- **Upload Only on Save**: Encrypted sync to cloud only after user confirmation, unselected solutions are immediately destroyed
- **Compliant with GDPR and domestic privacy regulations**
✔ **Model Lightweighting and Dynamicity**
- Self-developed Lighting Scene Model with pruning + quantization, **size < 3MB**
- Uses **On-Demand Loading + Online Hot Update** mechanism:
- Initial installation package doesn't contain the complete model, saving storage
- Supports OTA dynamic upgrades, effects can be optimized without releasing new versions
✔ **Intelligent Scene Adaptation**
- Model can recognize room types (living room/bedroom/dining room), device combinations (light strips/spotlights/ceiling lights), and other contexts
- Automatically generates professional lighting solutions with coordinated styles and distinct layers, such as:
- Warm Home Mode
- Focused Work Mode
- Romantic Dinner Mode
- Festival Atmosphere Mode (Spring Festival red, Christmas gold, etc.)
### Application Scenarios
| Application Scenario | Feature Description |
| ------------ | ------------------------------ |
| Home DIY Lighting Design | Users select intent tags like "Party" or "Reading", AI automatically generates whole-home lighting solutions |
| Smart Lighting Sales Demo | Sales staff generate multiple lighting effects on-site with one tap, improving customer conversion rates |
| Commercial/Hotel Rapid Deployment | Batch generation of unified lighting templates for multiple identical floor plans, saving construction and debugging time |
| Offline Emergency Scene Configuration | Can still create/switch lighting scenes without network, ensuring basic experience |
| Senior/Child-Friendly Interaction | No complex operations needed, simply click a style to apply, lowering usage barriers |
### Custom Model Application
If you are interested in training a custom model, visit the [On-App AI Custom Model Application Platform](https://developer.tuya.com/model/apply) to submit your request. The platform supports custom datasets, training and deployment to build tailored AI capabilities for your business scenarios.