Formal
Extract preview frames, then generate a polished caption.
AI-powered multi-style video captioning for short clips.
Built as an AMD Developer Hackathon Act II concept with a Fireworks AI-ready frontend flow: upload, sample frames, generate style variants, review, and copy.
Prototype Console
Waiting for a short clip
Video Upload Panel
No video selected
Choose a local video file to preview and sample frames.
Video preview appears here
Frame Extraction Preview
Base64 image data is stored in browser state for the future Fireworks AI captioning step.
Caption Generator Panel
Fireworks AI reads the sampled frames through a secure API route. Extract frames first; fallback mock captions keep the demo moving when the API is unavailable.
Extract preview frames, then generate a polished caption.
Waiting for frames before the wit engine clocks in.
Frame tensors pending. Caption runtime not yet initialized.
Upload, sample, and this card gets much funnier.
Visual summary appears after captions are generated.
Safety note appears after captions are generated.
Responsible AI Notes
Architecture
User selects a short clip and optional context.
Browser canvas captures representative preview frames.
Fireworks Vision reads sampled frames and user-provided context.
One base caption becomes four audience-ready styles.
Creator checks tone, accuracy, and sensitive context.
Approved captions are copied into the publishing workflow.