Simple Pics — AI Image Toolbox Mini Program
A live WeChat mini program bundling ~25 image tools in three groups. The AI suite — background removal, OCR, upscaling, style transfer, text-to-image and more — calls Tencent Hunyuan VLM through cloud functions only, so API keys never ship to the client. Everyday tools cover compress, crop, watermark and ID photos, and the creative section hides LSB steganography in plain sight. Heavy pixel work is hand-written on-device: a custom GIF encoder, perceptual hashing and EXIF handling on Canvas 2D.
Responsibilities
- Product & architecture design (subpackages: AI / GIF / tools)
- AI engineering: Hunyuan VLM via cloud functions, structured output & failure handling
- On-device imaging: custom GIF encoder, LSB steganography, perceptual hash, EXIF
- Launch & continuous iteration with a feature timeline
The AI Suite
Twelve AI capabilities: image description, captioning, background removal, style transfer, OCR, object eraser, upscaling, colorization, AI chat, outpainting, text-to-image and avatar generation.
- Tencent Hunyuan VLM (hunyuan-vision) is called only from cloud functions — keys stay server-side
- Structured outputs with validation, so model replies become usable fields
- Quota, timeout and fallback paths handled for a consumer audience
On-Device Imaging
- Hand-written GIF encoder (make / edit / batch) on Canvas 2D
- LSB steganography for hidden watermarks; perceptual hashing for similarity
- EXIF reading and writing via piexifjs
Architecture
- Three subpackages (pkgAi / pkgGif / pkgTools) keep the main bundle lean
- WeChat Cloud functions + storage; no separate backend to operate
Live on WeChat and actively iterated — a real testbed for shipping LLM/VLM capabilities inside consumer products.