[Feature] First & Last Frame Control - Keyframe Images for Scene Boundaries

[vzeman]

Summary

Implement control over the first and last frames of each video scene. Many AI video models support image-to-video generation where the start and/or end frame is specified — this feature unlocks seamless scene continuation and precise visual control.

Requirements

• First frame (start image) picker: upload from disk, take from generated video, AI-generate
• Last frame (end image) picker: same options as first frame
• "Generate start image" flow: select text-to-image model from Replicate, enter prompt, generate options (show 4 variations), pick one
• Auto-populate first frame from last frame of previous scene (continuation mode)
• First/last frame preview side-by-side in scene editor
• Crop/zoom tool to adjust image framing before use
• "Use last frame from video" extracts the final frame from a generated scene video
• "Use first frame from video" extracts the opening frame
• Frame extraction at custom timestamp (e.g., use frame at 2.5s as reference)
• Reference images section: upload multiple images for style transfer / consistency
• Image gallery per scene showing all uploaded/generated images
• Delete/replace images
• Image metadata display (dimensions, source model, generation prompt)

Acceptance Criteria

• Users can upload or generate a first/last frame for any scene
• Auto-populate correctly pulls last frame from previous scene video
• Frame extraction tool works on any generated video at any timestamp
• Crop/zoom tool adjusts framing before saving

Related

Part of the ai-video-editor Flutter application. Depends on #63 (Scene Editor). See also #66 (Text-to-Image Generation).