Media Studio is an AI toolkit that produces complete laptop review content for the Phát Lâm / Hưng Phát Laptop channel: a consistent set of product photos, banners and thumbnails, benchmark charts, AI models, Vietnamese voice-over and podcasts, and review videos built automatically from a single product link. Every feature can be called through the REST API /api/v1, so your article pipeline can run fully unattended.
Current version: v1.13.14 (2026-10-18) · Live: media.phatdev.net (account required) · See the Changelog.
Key points
- The real laptop stays untouched. Product photos keep the exact body, keyboard and ports - the AI only changes the setting, lighting and camera angle. Every image is automatically scored against the original photo and regenerated until it passes.
- From link to video. Paste a product link and the app reads the specs, looks up real benchmarks, writes the script, records the voice-over and renders a 1080p Remotion video with 3D motion.
- Free to run. Images are generated with GPT Image 2 (no separate API key needed); when the usage limit runs out it switches automatically to Nano Banana (Gemini). Upscaling, background removal and speech recognition all run locally.
- Ready for automation. The API uses key authentication, background jobs (
async), duplicate-generation protection on retries, and reports which engine produced each image.
What's new (Aug to Oct 2026)
| Area | Major changes |
|---|---|
| Video | Built with Remotion (11 React templates) instead of HyperFrames · the "pro" motion set (3D camera, flipping text) · text and cards appear exactly when the voice mentions them · CPU/GPU/Blender/game/office benchmark scenes with a plain-language "What this means" line · hook-style opening lines |
| Banner | Fully rebuilt as separate layers: AI background + original laptop + text · AI-drawn Featured images including the text, with an OCR spelling check before acceptance |
| AI engines | Images by GPT Image 2 · automatic fallback to Nano Banana when the usage limit runs out · text and image-reading tasks moved to Gemini · Grok removed |
| Image processing | Default upscaler UltraSharpV2 (beat Real-ESRGAN on 10 of 10 images) · video background removal now has a quality check and 3 fallback engines |
| Operations | 30-day storage (old files move to private storage and are pulled back when reopened) · fail-closed API with duplicate-generation protection · detailed API error logs · banner rendering no longer hangs |
Documentation index
Getting started
- Architecture and AI engines - the engines and how they back each other up.
Features
- Product photos - reshoot, background swap, automatic scoring, models, Brand Kit.
- Banners, thumbnails and charts - layered rendering, Featured images, benchmark charts.
- Automatic review videos - Remotion, benchmark scenes, voice sync.
- Voice-over and podcasts - Gemini TTS, VieNeu, transcription.
- Background removal and upscaling - rembg / macOS Vision / Photoshop, UltraSharpV2.
Integration
- The /api/v1 API for pipelines - authentication, background jobs, main endpoints.
- Operations and security - background running, remote access, self-cleaning storage, testing.
Reference
Screenshots
