🚀 gflow-api-proxy

Multi-profile HTTP API proxy for Google Flow (Veo 3.1 & Imagen / Nano Banana) with Playwright queue serialization, local media hosting and Cloudflare R2 support.

---

🎯 Features

• Queue Serialization Anti-Lock: Automatically coordinates Playwright Chrome instances per Google profile to prevent ProfileLockedError and singleton crashes.
• Local Media Server: All generated videos/images are stored locally in storage/ and served via /media/:filename with HTTP Range streaming support.
• Tailscale & Remote Ready: Auto-detects Tailscale IP to return direct remote URLs for browsing from your phone or remote laptop.
• Instant Markdown Preview: Returns markdown_preview and Base64 Data URI for immediate visual rendering inside chat UIs (like Paseo).
• Cloudflare R2 On-Demand: Pass "upload_r2": true to upload directly to R2 and get a public CDN link.
• OpenAI Drop-In Compatible: Standard /v1/images/generations and /v1/models endpoints.

---

🚀 Quick Start

Start Server

git clone https://github.com/cadorowo/gflow-api-proxy.git
cd gflow-api-proxy
npm install
npm start

By default, listens on port 1235 (e.g. http://0.0.0.0:1235).

Health Check

curl http://localhost:1235/health

---

📡 API Endpoints

1. Video Text-to-Video (Veo 3.1)

curl -X POST http://localhost:1235/v1/video/t2v \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "Cinematic slow push-in shot of ocean waves at sunrise",
    "model": "omni-flash",
    "aspect": "9:16",
    "wait": true
  }'

2. Video Image-to-Video

curl -X POST http://localhost:1235/v1/video/i2v \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "The subject turns head right with a subtle smile",
    "initial_frame": "/path/to/frame.jpg",
    "aspect": "9:16"
  }'

3. Image Text-to-Image (Imagen / Nano Banana)

curl -X POST http://localhost:1235/v1/image/t2i \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "Minimalist geometric 3D character with headphones, dark aesthetic",
    "model": "nano-pro",
    "aspect": "1:1"
  }'

4. OpenAI Drop-In (/v1/images/generations)

curl -X POST http://localhost:1235/v1/images/generations \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "Futuristic cyberpunk vehicle",
    "size": "1024x1024"
  }'

5. Async Mode & Polling

Add "wait": false to any generation request to get immediate 202 Accepted with task_id:

curl http://localhost:1235/v1/tasks/<task_id>