FlyReq Image Studio

简体中文 | English

Overview

FlyReq Image Studio is a self-hosted AI image and video creation workspace for individuals and teams. It combines a static Next.js 16 + React 19 PWA frontend with a lightweight Node.js, SQLite, and WebSocket backend that queues jobs and proxies image- and video-generation APIs.

Built from tianjiangqiji/nova-image-studio and maintained at doudou770/flyreq-image-studio.

Highlights:

• Multi-model access: Configure image, video, and text models independently across Google, OpenAI, xAI, and New API.
• Capability-aware controls: Show resolution, references, duration, aspect ratio, and advanced options based on each model's limits.
• Unified image and video creation: Generate images, edit references, create GIFs and videos, and work on an infinite canvas.
• Production-friendly deployment: Configure branding, defaults, concurrency, and rate limits through environment variables without overwriting existing user settings.
• Fast setup from links: Prefill image, video, or text models from a URL, then save only after user confirmation.
• Reliable, traceable jobs: Persistent queues, live WebSocket updates, polling fallback, failure diagnostics, and local backup are built in.

Product Preview

Image Workspace

Wide	Mobile

Video Workspace

Wide	Mobile

Agent Mode

Planning	Generation

GIF Workflow

Generation	Refinement

Infinite Canvas

Prompt Optimization

Entry point	Result

Inspiration and Assets

Prompt Gallery	My Assets

Configuration and Creation

Reverse Prompt	Settings

Sponsor

Thanks to FlyReq for sponsoring this project. FlyReq is an AI model API relay platform focused on high-discount access for developers and teams looking for a cost-effective way to connect the models they need. New registrations receive trial credit for validating model capabilities and integration flows. Visit flyreq.com to learn more and get started.

Supported Image and Video Models

Workflows

Deployment

# 1. Create and enter the deployment directory
sudo mkdir -p /opt/fis
cd /opt/fis

# 2. Download the Docker Compose configuration
sudo curl -fsSL \
  https://raw.githubusercontent.com/doudou770/flyreq-image-studio/master/docker-compose.yml \
  -o docker-compose.yml

# 3. Download the environment template as .env
sudo curl -fsSL \
  https://raw.githubusercontent.com/doudou770/flyreq-image-studio/master/backend/.env.example \
  -o .env

# 4. Download prompt and blacklist configuration
sudo curl -fsSL \
  https://raw.githubusercontent.com/doudou770/flyreq-image-studio/master/backend/prompts.json \
  -o prompts.json
sudo curl -fsSL \
  https://raw.githubusercontent.com/doudou770/flyreq-image-studio/master/backend/blacklist.json \
  -o blacklist.json

# 5. Create the persistent data directory
sudo mkdir -p data

# 6. Edit configuration when needed
sudo nano .env

# 7. Start the service
sudo docker compose up -d

image: ghcr.io/doudou770/flyreq-image-studio:latest

echo YOUR_GITHUB_TOKEN | docker login ghcr.io -u YOUR_GITHUB_USERNAME --password-stdin

/opt/fis/
├── docker-compose.yml
├── .env
├── prompts.json
├── blacklist.json
└── data/

FLYREQ_TASK_DB: /app/backend/data/flyreq-tasks.sqlite
FLYREQ_IMAGE_DIR: /app/backend/data/flyreq-images

sudo docker network create 1panel-network

cd /opt/fis
sudo docker compose restart

FLYREQ_BASE_URL_REWRITE_MAP={"https://flyreq.com":"http://new-api:3000"}

FLYREQ_BASE_URL_REWRITE_MAP={"https://flyreq.com":"http://new-api:3000","https://api.example.com":"http://example-new-api:3000"}

cd /opt/fis
sudo docker compose pull
sudo docker compose up -d --force-recreate

npm ci
npm run build

frontend/out/
backend/server.js
backend/package.json
backend/package-lock.json
backend/prompts.json
backend/blacklist.json
backend/.env

npm ci --omit=dev
npm start

npm run go

git clone https://github.com/doudou770/flyreq-image-studio.git
cd flyreq-image-studio
npm install
cp backend/.env.example backend/.env
# Windows: Copy-Item backend/.env.example backend/.env
npm run dev

npm run dev:frontend   # Next.js dev server only (HMR; no static export)
npm run dev:backend    # backend server.js only
npm run build          # static frontend output in frontend/out/
npm start              # backend server.js
npm run lint           # frontend ESLint
npm test               # Vitest watch mode
npm run test:run       # Vitest once
npm run go             # build and package root out.zip

docker build -t flyreq-image-studio:latest .
docker tag flyreq-image-studio:latest ghcr.io/doudou770/flyreq-image-studio:latest
docker push ghcr.io/doudou770/flyreq-image-studio:latest

Supported Models and Protocols

Presets define a safe capability boundary, not a provider lock-in. Supply the actual base URL, model ID, and API key for a compatible service. Google and xAI image APIs do not receive stream=true; OpenAI Images-compatible GPT Image 2 requests can enable streaming by default.

Why It Is Different

• Intent-aware Agent routing: The Agent considers the requested resolution, available image models, and reference-image aspect ratio, then normalizes settings to the selected model's supported range.
• One configuration surface: Deployment variables provide a branded first-run experience, while external links can hand users a model draft that still requires confirmation.
• Compatibility with diagnostics: Server-side base-URL rewrites route public settings to internal services without changing what users saved. Upstream response bodies are retained for investigation.
• Recoverable job system: SQLite-backed jobs, WebSocket updates, reconnect with polling fallback, on-disk outputs, retry, download, backup, and restore are built in.

Prompt Gallery

PROMPT_GALLERY_MODE controls how the gallery is exposed:

• 1: Always visible.
• 2: Private, protected by PROMPT_GALLERY_PASSWORD.
• 3: Hidden.

Gallery content lives in backend/prompts.json and supports filtering through backend/blacklist.json.

External Model Configuration Links

External sites can link to FlyReq Image with a provider query parameter containing one image, text, or video model configuration as JSON. The application opens Settings, fills the corresponding model draft, removes all configuration parameters from the address bar, and waits for the user to save. Importing a link never writes configuration automatically.

Use URL-encoded JSON in production. The raw links below are readable examples; generate a production value with encodeURIComponent(JSON.stringify(payload)).

Image model example

{
  "type": "image",
  "preset": "gpt-image-2",
  "provider": "openai",
  "modelKey": "flyreq-gpt-image-2",
  "name": "FlyReq",
  "modelId": "gpt-image-2",
  "baseUrl": "https://flyreq.com",
  "apiKey": "YOUR_API_KEY",
  "maxRefImages": 16,
  "maxOutputSize": "4K",
  "supportsTemperature": false,
  "streamImages": true
}

https://image.flyreq.com/en/?provider={"type":"image","preset":"gpt-image-2","provider":"openai","modelKey":"flyreq-gpt-image-2","name":"FlyReq","modelId":"gpt-image-2","baseUrl":"https://flyreq.com","apiKey":"YOUR_API_KEY","maxRefImages":16,"maxOutputSize":"4K"}

URL-encoded JSON:

https://image.flyreq.com/zh/?provider=%7B%22type%22%3A%22image%22%2C%22preset%22%3A%22gpt-image-2%22%2C%22provider%22%3A%22openai%22%2C%22modelKey%22%3A%22flyreq-gpt-image-2%22%2C%22name%22%3A%22FlyReq%22%2C%22modelId%22%3A%22gpt-image-2%22%2C%22baseUrl%22%3A%22https%3A%2F%2Fflyreq.com%22%2C%22apiKey%22%3A%22YOUR_API_KEY%22%2C%22maxRefImages%22%3A16%2C%22maxOutputSize%22%3A%224K%22%7D

When complete, the imported image model becomes the text-to-image and image-to-image default.

Text model example

{
  "type": "text",
  "provider": "openai",
  "modelKey": "flyreq-text-default",
  "name": "FlyReq Text",
  "modelId": "gpt-5.4-mini",
  "baseUrl": "https://flyreq.com",
  "apiKey": "YOUR_API_KEY",
  "note": "OpenAI Responses-compatible text model"
}

https://image.flyreq.com/en/?provider={"type":"text","provider":"openai","modelKey":"flyreq-text-default","name":"FlyReq Text","modelId":"gpt-5.4-mini","baseUrl":"https://flyreq.com","apiKey":"YOUR_API_KEY","note":"OpenAI Responses-compatible text model"}

When complete, the imported text model becomes the initial default for Agent, reverse prompt, prompt optimization, and image description. Users can change each default before saving. Text models support openai and google.

Video model example

{
  "type": "video",
  "protocol": "openai",
  "modelKey": "flyreq-video-default",
  "name": "FlyReq Video",
  "modelId": "sora-2",
  "baseUrl": "https://api.openai.com",
  "apiKey": "YOUR_API_KEY"
}

https://image.flyreq.com/en/?provider={"type":"video","protocol":"openai","modelKey":"flyreq-video-default","name":"FlyReq Video","modelId":"sora-2","baseUrl":"https://api.openai.com","apiKey":"YOUR_API_KEY"}

When complete, the imported video model becomes the video-generation default. Video links accept the new-api, openai, and xai protocols. Use the explicit protocol field for new links; the historical provider=openai form remains mapped to the legacy video endpoint.

All standard backend logs are mirrored to a daily JSONL file while remaining available on the console. Files are written to backend/logs/application/application-YYYY-MM-DD.log and include server startup, task queues, image/video processing, cleanup, and WebSocket events. Application file logging is enabled by default; set FLYREQ_FILE_LOG_ENABLED=false to disable it or use FLYREQ_LOG_DIR to change the directory.

Every image-upstream generation request and response for OpenAI, xAI, or Google is also emitted as a structured [image-upstream] log and appended to a daily JSONL file at backend/logs/image-upstream/image-upstream-YYYY-MM-DD.log using the process-local date. The log includes the method, URL, headers, body, HTTP status, response headers/body, and context such as the model, mode, output size, and aspect ratio. API keys, authorization headers, cookies, and signed-URL query parameters are sanitized. Request data URLs and multipart images are replaced with MIME or byte-count summaries. OpenAI b64_json response values are replaced with the fixed b64_json_value placeholder, while Gemini inlineData.data values use byte-count summaries, so raw image data is never written to logs. Logging is enabled by default and can be disabled with FLYREQ_IMAGE_UPSTREAM_LOG_ENABLED=false. Ordinary response bodies are limited to 65,536 characters by default and can be adjusted with FLYREQ_IMAGE_UPSTREAM_LOG_MAX_CHARS; the directory can be changed with FLYREQ_IMAGE_UPSTREAM_LOG_DIR.

Every video-upstream request and response for creation, polling, and result download is also written as a structured [video-upstream] log and appended to a daily JSONL file at backend/logs/video-upstream/video-upstream-YYYY-MM-DD.log using the process-local date. Every stage record includes the local taskId, display model name, upstream model ID, resolution, and current elapsed time. A final task-summary record captures the terminal status and exact totalDurationMs when a task completes, fails, or is cancelled. The log also includes the stage, method, URL, HTTP status, headers, request parameters, response body, and task context. API keys, authorization headers, cookies, and signed-URL query parameters are sanitized; data URLs and multipart media are reduced to type, name, and byte-count metadata. Responses with a video/* content type are represented only by a media-type and byte-count placeholder, so video bodies are never written to logs. Logging is enabled by default and can be disabled with FLYREQ_VIDEO_UPSTREAM_LOG_ENABLED=false. Ordinary response bodies are limited to 65,536 characters by default and can be adjusted with FLYREQ_VIDEO_UPSTREAM_LOG_MAX_CHARS; the directory can be changed with FLYREQ_VIDEO_UPSTREAM_LOG_DIR. Docker Compose persists /app/backend/logs under host ./logs and defaults daily rotation to the Asia/Shanghai timezone.

Fields and behavior

Matching first uses modelKey, then name + modelId + baseUrl; otherwise, a new draft is added. Incomplete payloads are retained as inactive drafts so the user can add missing fields. Legacy configureModel=1&type=... multi-parameter links remain supported for all three types, and every recognized parameter is removed after parsing.

The API key is briefly present in browser history, proxy logs, chat previews, and referrer metadata before the application cleans the URL. Prefer short-lived keys or omit apiKey and let the user enter it locally.

Task System

• Jobs enter a server-side queue with configurable concurrency and rate limits.
• Browsers receive job and queue changes through WebSocket, reconnect automatically, and fall back to HTTP polling after repeated failures.
• Generated files are stored on disk and served from /api/flyreq/images/:taskId/:index.
• Jobs expire after 12 hours and are cleaned up automatically.
• A restart marks incomplete jobs as failed and removes their partial output, preventing orphaned tasks.

Engineering and UX

• Installable PWA through next-pwa.
• Responsive desktop, tablet, and mobile layouts.
• Light and dark themes plus wide and narrow workspace layouts.
• Task history persisted with IndexedDB and localStorage.
• One-click backup and restore through JSZip, including partial recovery when older settings are incompatible.
• Virtualized lazy loading for historical images through @tanstack/react-virtual.
• Random images, toast notifications, and confirmation dialogs.

Important Environment Variables

For example, keep a public URL in browser-side model settings while routing server-side requests to a Docker network service:

FLYREQ_BASE_URL_REWRITE_MAP={"https://flyreq.com":"http://new-api:3000"}

This mapping changes only outbound server requests. It never rewrites the user's stored model configuration.

Most runtime settings in .env take effect without a restart, including concurrency, rate limits, queue capacity, accepting-new-tasks, outbound base-URL rewrites, Prompt Gallery settings, and the image-model key guide. Restart after changing startup settings such as PORT, HOSTNAME, or NODE_ENV.

Video Protocol Configuration

The default video model uses FLYREQ_DEFAULT_VIDEO_MODEL_PROTOCOL=openai and FLYREQ_DEFAULT_VIDEO_MODEL_MODEL_ID=sora-2. Supported values are:

• new-api: New API generic video protocol, using POST /v1/video/generations.
• openai: OpenAI Videos (Sora), using POST /v1/videos. This is the default.
• xai: xAI Videos, using POST /v1/videos/generations.

When changing the protocol, also configure FLYREQ_DEFAULT_VIDEO_MODEL_BASE_URL and FLYREQ_DEFAULT_VIDEO_MODEL_MODEL_ID for that upstream service. Protocol capabilities are defined in backend/video-protocol-capabilities.json and can be overridden through FLYREQ_VIDEO_PROTOCOL_CONFIG_OVERRIDES.

API

Task statuses

• queued: Waiting for scheduling.
• processing: Calling the upstream API.
• completed: Succeeded; result.images contains result URLs.
• failed: Failed; see error.
• expired: Exceeded the task TTL.

Troubleshooting

Why not use next start in production?

The project uses output: 'export', producing a static frontend/out/ directory. server.js hosts both the static files and task API, so next start is not used.

Can I deploy only frontend/out/?

The UI opens, but task submission, Agent, and history synchronization require /api/flyreq/*; run server.js as well.

Does the database need a separate backup?

The service creates it automatically. To retain task data, back up flyreq-tasks.sqlite with its WAL/SHM files and flyreq-images/. A restart marks remaining active tasks as failed and cleans up their output.

How do I temporarily stop accepting new tasks without stopping the service?

Set the following in .env:

FLYREQ_ACCEPT_NEW_TASKS=false

It takes effect after saving. Wait for in-flight tasks before restarting for an upgrade. Set it to true or leave it empty to accept tasks again.

When do tasks expire?

Tasks expire 12 hours after creation. When the frontend receives a result, it calls /ack to extend retrieval by two minutes. The server removes the database record and result images after expiry.

Why can an upstream console show success while FlyReq Image reports 504?

Cloudflare, Nginx, or another gateway can close a long-running response before the upstream image job finishes. Prefer a Docker-internal or DNS-only upstream address, configure FLYREQ_BASE_URL_REWRITE_MAP, and enable streaming image requests for compatible OpenAI Images endpoints. The original upstream error is kept in the task failure message.

Acknowledgements

The infinite-canvas workspace is built on infinite-canvas. Thanks to its author, basketikun, for the open-source contribution, and to tianjiangqiji for the open-source UI.

Thanks to the Linux.do community for its support.

Star History

License

Released under the AGPL-3.0 license.

For the primary Chinese documentation, complete deployment guide, environment-variable reference, and operational FAQ, see README.md.