Spaces:
Runtime error
Runtime error
| # Pitch Deck Outline | |
| Use this as the slide plan for the required presentation deck. | |
| ## Slide 1 - Title | |
| ElevenClip.AI | |
| AI clip studio for turning long-form videos into personalized short-form clips. | |
| Include: | |
| - AMD Developer Hackathon | |
| - Track 3 - Vision & Multimodal AI | |
| - GitHub URL | |
| - Hugging Face Space URL | |
| ## Slide 2 - Problem | |
| Long-form creators need short-form distribution, but editing clips manually is slow. | |
| Key points: | |
| - Two-hour videos can take hours to review. | |
| - Good clips depend on audience, niche, tone, and platform. | |
| - Subtitles and vertical export add repetitive work. | |
| ## Slide 3 - Solution | |
| ElevenClip.AI automates the first editing pass. | |
| Workflow: | |
| Video input -> Whisper transcript -> Qwen highlight scoring -> ffmpeg clip rendering -> human review/editor -> downloads | |
| ## Slide 4 - Product Demo | |
| Show screenshots or short GIFs of: | |
| - Channel profile | |
| - Pipeline progress | |
| - Transcript/highlights | |
| - Clip editor | |
| - Approved/downloaded clips | |
| ## Slide 5 - AI Architecture | |
| Model roles: | |
| - Whisper Large V3: multilingual transcription, including Thai. | |
| - Qwen2.5-7B-Instruct: profile-aware highlight detection. | |
| - Qwen2-VL-7B-Instruct: visual reactions, scene changes, and on-screen text. | |
| - ffmpeg: subtitle burn-in and platform export. | |
| ## Slide 6 - AMD + ROCm | |
| Why AMD matters: | |
| - Long videos need high-throughput inference. | |
| - MI300X memory helps with large models and long transcripts. | |
| - ROCm + PyTorch enables Whisper inference. | |
| - vLLM ROCm enables faster Qwen serving. | |
| ## Slide 7 - Benchmark | |
| Replace placeholders after cloud credits arrive. | |
| | Run | Hardware | Total Time | Clips | | |
| | --- | --- | ---: | ---: | | |
| | CPU baseline | CPU | TBD | 10 | | |
| | AMD GPU | MI300X + ROCm | TBD | 10 | | |
| Goal: 2-hour video -> 10 subtitled clips in under 10 minutes on MI300X. | |
| ## Slide 8 - Business Value | |
| Target users: | |
| - YouTubers | |
| - Podcasters | |
| - Educators | |
| - Streamers | |
| - Agencies | |
| - Brand marketing teams | |
| Value: | |
| - Save editing time. | |
| - Increase short-form output. | |
| - Keep creator control. | |
| - Support multilingual creators. | |
| ## Slide 9 - What We Built | |
| Current MVP: | |
| - FastAPI backend | |
| - React editor | |
| - YouTube/upload input | |
| - Demo pipeline | |
| - Clip rendering and subtitles | |
| - Hugging Face Space | |
| - AMD deployment plan | |
| Next: | |
| - Real Whisper + Qwen on MI300X | |
| - Qwen2-VL frame analysis | |
| - Benchmark table | |
| - Better subtitle styling presets | |