---
title: "Media Studio"
url: https://phatdev.net/en/projects/media-studio
lang: en
alternate:
  vi: https://phatdev.net/vi/projects/media-studio
markdown: https://phatdev.net/en/projects/media-studio.md
---

# Media Studio

> AI app for laptop review content: from one product link it produces true-to-device photo sets, spell-checked Vietnamese banners and voiced review videos with real benchmarks, all callable through a REST API for the article pipeline.

- Role: AI · Full-stack · Content automation
- Status: In daily use for review content
- Technology: Node.js, Remotion, GPT Image 2, Gemini, FFmpeg
- Live site: https://media.phatdev.net
- Docs and changelog: https://phatdev.net/docs/media-studio
- All projects: https://phatdev.net/en/projects.md

## Project facts

- Role: Design, development, operations
- Engines: GPT Image (Codex), Gemini, VieNeu, TinyPNG
- Version: 1.15.23
- Outputs: Photos, banners, scripts, voiceovers, videos

## Metrics

- 4.721: images generated in the library
- 9: tools in one dashboard

## Highlights

- **Quick create from one link**: Paste a product link to get a narrated, subtitled review video, a photo set, banners or a podcast.
- **Consistent product photos**: Laptop photo sets kept consistent per scene (studio, gaming, office...), with chosen angles and variant counts.
- **Image processing**: Background removal with macOS Vision to transparent PNG, UltraSharp upscaling to 2K, 4K or 8K.
- **Scripts and voice**: Review and comparison scripts, Gemini or VieNeu voiceover, two-host podcasts.
- **AI models and banners**: Consistent uniformed AI models, cover banners, thumbnails, laptop and component comparisons.

## In-depth analysis

### The problem

A single laptop review needs a lot: a consistent product photo set, a cover image, comparison banners, benchmark charts, a script, voiceover and a video. Each lives in a different tool, AI tends to redraw device details (keycaps, ports, logos), Vietnamese text on images loses its diacritics, and benchmark numbers are looked up by hand.

### The solution

Media Studio puts the whole workflow into one web app running on a home machine, usable by hand or through an API. The article pipeline sends a product link and gets photos, banners and a video back.

#### True-to-device product photos
Reshoot keeps the original camera angle, background swap keeps every device pixel, and AI models stay consistent. Each image is scored by AI against the reference (chassis, keycaps, ports) and regenerated until it passes; studio shots went from a match of 72 to 89-92.

#### Layered banners
AI paints the background and headline, the original laptop photo is composited on top. Vietnamese text is read back by OCR; a single wrong diacritic triggers a retry or a switch to HTML text.

#### Review videos from one link
Real benchmarks (NotebookCheck, Blender Open Data), a script that opens with a hook, Gemini voiceover, and a 1080p Remotion render with 11 scene templates and 3D camera moves. Whisper word timings make each spec appear exactly when the voice mentions it.

#### Image and audio processing
Multi-engine background removal with quality checks, on-device UltraSharpV2 upscaling, TinyPNG compression, Gemini and VieNeu voices, two-host podcasts, local speech recognition.

#### Fallback engines and API
GPT Image 2 (Codex) generates images and falls back to Nano Banana when the quota runs out; Gemini 3.8 Flash handles writing and image reading. The REST `/api/v1` is key-protected, with background jobs, duplicate-safe retries and a quota-aware health endpoint.

### Compared to popular tools

| Feature | Media Studio | Canva | Photoroom |
|---|---|---|---|
| Background removal, product upscaling | Yes | Yes | Yes |
| Consistent sets that keep device details | Yes | Partial | Partial |
| Automatic AI-vs-reference scoring | Yes | No | No |
| Vietnamese-text banners with OCR checks | Yes | Partial | No |
| Review video from a link, voiced, real benchmarks | Yes | Partial | No |
| Multiple AI engines with automatic fallback | Yes | No | No |
| REST API for pipelines | Yes | Partial (paid) | Partial |

> Not a Canva replacement for free-form design. Media Studio excels at **automating product review content through a pipeline and API**, running almost free on a home machine.

## License

Source code is not public, all rights reserved. The code of this project is not kept on the web server, so there is no license file to check against.

See the License and credits page: https://phatdev.net/en/licenses.md