title: ElevenClip.AI
emoji: 🎬
colorFrom: green
colorTo: yellow
sdk: static
app_file: index.html
pinned: false
short_description: AI studio for short-form video clips.
tags:
- amd
- rocm
- mi300x
- qwen
- whisper
- huggingface
- video-ai
- multimodal
ElevenClip.AI
ElevenClip.AI is an AI clip studio that turns long-form videos into personalized short-form clips for TikTok, YouTube Shorts, and Instagram Reels.
This Space is the public demo page for our AMD Developer Hackathon project on lablab.ai.
Hackathon Track
Track 3 - Vision & Multimodal AI
ElevenClip.AI processes video, audio, text, and subtitle outputs:
- Audio: Whisper Large V3 transcription through Hugging Face.
- Text: Qwen2.5 highlight scoring from timestamped transcripts.
- Video: optional Qwen2-VL visual analysis for reactions, scene changes, and on-screen text.
- Rendered media: ffmpeg exports vertical short-form clips with subtitles.
Why It Matters
Creators often publish long videos but still need short clips for discovery platforms. Finding the best moments, trimming clips, writing subtitles, and exporting vertical videos can take hours.
ElevenClip.AI automates the first pass:
- Upload a video or paste a YouTube URL.
- Transcribe the content with Whisper.
- Use Qwen to identify high-engagement highlight moments.
- Render short-form clips with subtitles.
- Let the creator trim, edit subtitles, approve, regenerate, and download final clips.
AMD + ROCm Plan
The production backend is designed for:
- AMD Developer Cloud
- AMD Instinct MI300X
- ROCm 6.x
- PyTorch with ROCm support
- vLLM ROCm backend for Qwen inference
Current local demo mode validates the upload, clipping, subtitle rendering, and human editing workflow. Real Whisper and Qwen inference will be enabled on AMD Developer Cloud once MI300X credits are active.
Links
- GitHub: https://github.com/JakgritB/ElevenClip.AI
- Hackathon: https://lablab.ai/ai-hackathons/amd-developer
Status
- Local MVP: working
- Hugging Face Space: landing page
- AMD Cloud credits: requested
- Real MI300X benchmark: pending