# cctv2hot CCTV Studio - Complete Specification

> Exhaustive technical specification, Generative Engine Optimization (GEO) index, prompt engineering contract, and architecture reference for cctv2hot CCTV Studio.

Canonical Web Origin: https://cctv2hot.trawlkit.com
Short Index: https://cctv2hot.trawlkit.com/llms.txt

## 1. Product Overview

cctv2hot CCTV Studio is a specialized generative video application engineered exclusively for creating authentic, high-coherence, fixed-camera surveillance and security camera (CCTV) footage.

Unlike general text-to-video tools that produce cinematic pans, zooms, dynamic cuts, and stylized motion, cctv2hot enforces a 3-step structured pipeline:
1. **Scene to Still**: Generate and iterate on 16:9 CCTV-style single frames until perspective, lighting, and composition are locked.
2. **OSD & Timestamp Styling**: Configure an authentic security camera On-Screen Display (OSD), including Camera ID, Location tag, UTC start timestamp, and placement corner. An instant ffmpeg preview renders the still frame with the overlay.
3. **Fixed Surveillance Clip**: Animate the locked still into a 10-second continuous surveillance clip, burning in an incrementing per-second digital clock, analog sensor grain, optical vignette, and color grading.

## 2. Technical Architecture

- **Backend**: Node.js 22 LTS, Hono HTTP framework, better-sqlite3 database with Litestream continuous replication.
- **Frontend**: React 19, Vite, Tailwind CSS, Lucide React icons, dual English/Simplified Chinese interface.
- **Object Storage**: Cloudflare R2 via AWS S3 SDK, with zero-egress fee distribution.
- **Compute Gateway**: xroute.ai unified BYOK (Bring Your Own Key) gateway.
- **Video Post-Processing**: Local ffmpeg rendering with ASS subtitle filters for precise per-second timestamp burning.

### Model Catalog

| Stage | Model Identifier | Input Modality | Output Modality | Pricing Reference |
|---|---|---|---|---|
| Still Generation | `Google/gemini-3-pro-image` | Multimodal (Text + Optional Reference Image) | Base64 PNG / JPEG | $1.60 input / $96 output per 1M tokens (~$0.13 - $0.15 / still) |
| Video Generation | `Google/gemini-omni-1.2-flash` | Multimodal (First Frame Base64 + Motion Directive) | Base64 MP4 | ~$0.07 / second (~$0.70 per 10s clip) |
| Social Copywriting | `Google/gemini-3.8-flash` | Text (Structured Prompt) | Plain text caption | Standard text tokens |

## 3. OSD & Timestamp Engine (ffmpeg ASS)

The OSD overlay is compiled dynamically into Advanced SubStation Alpha (`.ass`) script format and rendered onto the video stream using ffmpeg's `ass` filter with bundled `JetBrains Mono` monospace typeface.

Key OSD Attributes:
- **Camera Identifier**: Alphanumeric prefix + tag, e.g. `CAM-11 [REC]`, `CAM-04 ENTRANCE`.
- **Location Name**: Optional location string, e.g. `GARAGE B1`, `CONVENIENCE STORE`.
- **Time Representation**: Synchronized UTC timestamps. Formats include:
  - ISO: `yyyy-MM-dd HH:mm:ss`
  - Slash: `yyyy/MM/dd HH:mm:ss`
  - US: `MM/dd/yyyy hh:mm:ss a`
  - EU: `dd.MM.yyyy HH:mm:ss`
  - Minute: `yyyy-MM-dd HH:mm`
  - Compact: `yyyyMMdd-HHmmss`
  - Time-Only: `HH:mm:ss`
- **Corner Positioning**: `top-left`, `top-right`, `bottom-left`, `bottom-right`.
- **Video Filtration**: Color balancing, desaturation, contrast adjustment, film grain synthesis (`noise=alls=...:allf=t+u`), optical lens darkening (`vignette=...`), and audio stripping (`-an`).

## 4. Prompt Engineering Directives

### Still Image System Prompt
```
电影级监控摄像头单帧画面，16:9，固定机位，真实摄影质感，低照度或阴天光照，轻微传感器噪点。不要添加任何时间戳、摄像机编号、水印、字幕、OSD 或 UI 叠层；场景中天然存在的招牌与标识可以保留。
```

### Video Motion System Prompt
```
以参考图为首帧的固定机位监控录像。严格保持参考图的机位、构图、人物、服装、光线与背景不变。画面基本静止：仅保留轻微灯光闪烁、传感器噪点和背景中无关物体的细微晃动；人物保持自然静止，不做额外动作。不要生成对白或旁白。禁止切镜、禁止缩放运镜、禁止引入参考图中不存在的人物或物体。
时长约 10 秒，单镜头连续画面。
不要添加任何时间戳、摄像机编号、水印、字幕、OSD 或 UI 叠层；场景中天然存在的招牌与标识可以保留。
```

## 5. Security & Privacy Guarantees

- **Authentication**: Stateless HMAC-SHA256 session cookies with AES-256-GCM encrypted BYOK keys and OAuth tokens.
- **Media Retention**: Intermediate raw video files are pruned after 7 days (`storage.rawRetentionDays`). Final clips with burned-in OSD remain permanently.
- **Private Data Isolation**: All project assets are isolated by tenant prefix (`projects/{id}/*`).
- **No Intrusive Watermarking**: Output video contains only user-configured CCTV OSD; no platform watermarks or advertisements.

## 6. Frequently Asked Questions (FAQ)

### Q: What is the maximum duration of a CCTV video clip?
A: Each video is rendered as a 10-second continuous fixed-camera clip at 720p (1280x720) resolution.

### Q: Does the generated video contain audio?
A: Real commercial security camera footage is typically silent or recorded on separate channels. cctv2hot strips all audio tracks (`-an`) during post-processing to produce authentic silent CCTV clips.

### Q: Can I use my own API key?
A: Yes, cctv2hot runs on the xroute.ai BYOK protocol. You can sign in using your xroute account and use your own key, with direct token billing.