Photo Liftoff

Photo Liftoff turns still photos into short 5-10 second video-generation jobs. It includes:

• a drag-and-drop photo workspace
• serious and funny prompt helpers
• duration controls for 5-10 second clips
• local mock generation for development
• a configurable Replicate-compatible AI video provider
• Railway deployment config

Run Locally

npm install
npm run dev

In another terminal, run the API:

npm run dev:api

Open http://127.0.0.1:5173.

Real AI Video Generation

The app works without secrets in AI_VIDEO_PROVIDER=mock mode. To call a real provider, set:

AI_VIDEO_PROVIDER=replicate
REPLICATE_API_TOKEN=your_token
REPLICATE_MODEL=owner/model-name

Some image-to-video models use different input names, so configure them without code changes:

VIDEO_PROMPT_FIELD=prompt
VIDEO_IMAGE_FIELD=start_image
VIDEO_DURATION_FIELD=duration
VIDEO_ALLOWED_DURATIONS=5,10
VIDEO_EXTRA_INPUT_JSON={"aspect_ratio":"16:9"}

For kwaivgi/kling-v2.1, the app automatically uses start_image and valid durations 5,10 if you leave VIDEO_IMAGE_FIELD unset. Other models may require different field names.

You can also use REPLICATE_VERSION instead of REPLICATE_MODEL for version-based predictions.

Deploy To Railway

1. Create a new Railway service from this GitHub repo.
2. Add the environment variables from .env.example.
3. Keep AI_VIDEO_PROVIDER=mock for smoke tests, or set AI_VIDEO_PROVIDER=replicate plus provider credentials.
4. Railway will run npm run build and npm start.

API

POST /api/prompts

Returns generated serious and funny prompts.

POST /api/videos

Multipart form fields:

• photo: image file
• variant: serious or funny
• duration: number from 5 to 10
• subject: optional subject hint
• motion: optional motion hint
• style: optional style hint

GET /api/videos/:id

Returns a mock job or the provider prediction status.