[Feature] AI Video Generation - Replicate API Integration with All Supported Models

[vzeman]

Summary

Integrate the Replicate API for AI video generation, supporting all available video generation models. This is the core generation engine of the application — every video clip produced by the editor flows through this component.

Requirements

• Replicate API client with API key configuration
• Supported models (and dynamically discover new ones):
  • Stable Video Diffusion (SVD, SVD-XT)
  • Kling (1.0, 1.5, 2.0)
  • Wan 2.1 (T2V, I2V)
  • Hailuo MiniMax
  • LTX Video
  • CogVideoX
  • AnimateDiff
  • Mochi 1
  • Ray (Luma Dream Machine)
  • Runway Gen-3
• Model discovery: fetch available models from Replicate with video tag
• Queue management: submit generation, poll for completion
• Generation status tracking: queued, processing, succeeded, failed, cancelled
• Cancel in-flight generation
• Retry failed generation with same or modified params
• Download and cache generated video locally
• Generation history per scene (all attempts with params and output)
• Side-by-side comparison of multiple generation attempts
• Estimated generation time display
• Cost tracking (Replicate billing per-second)
• Batch generation: generate all pending scenes sequentially or in parallel (configurable concurrency)
• Webhook support (optional) for completion notifications

Acceptance Criteria

• Replicate API client authenticates and lists available video models
• User can submit a generation request and see live status updates
• Generated video is downloaded and cached locally
• Failed generations can be retried
• Generation history persists across app sessions
• Batch generation processes multiple scenes concurrently

Related

Part of the ai-video-editor Flutter application. See also #75 (Cost Tracking & API Key Management).