Back to Build Matcha & Code Gallery
Project spotlight
Approved
Shortify_ads
by shortifyAds
Create, analyze, and remix videos with a multimodal AI workspace.
About This Project
The product combines three workflows in one responsive dashboard:
- Text to Video Users enter a prompt, which is refined into a stronger cinematic generation prompt using Kimi on GMI Cloud, then passed to PixVerse 5.6 via GMI to generate video.
- Long Video to Best Clips Users upload an MP4 or provide a YouTube link. The system preprocesses the video, analyzes it with a multimodal model, ranks the strongest moments, and extracts the best short clips. This helps turn long-form content into Shorts, Reels, or highlight segments.
- Multimodal Guided Generation Users can combine text, images, and video as input. The system analyzes all modalities together, creates a structured generation plan, and produces better segment-level outputs for stitched final videos. The platform is designed as a premium chat-like workspace rather than a basic form-based tool. Users can upload assets, track live job progress, review outputs, iterate on generations, and return later with their history restored. To make the system more context-aware over time, HydraDB stores semantic memory such as user preferences, prompt lineage, summaries of prior inputs and outputs, and reusable creative direction. Redis handles live job-state tracking, Neon Postgres stores durable product data and history, and Cloudflare R2 is used for temporary media storage with aggressive cleanup to reduce cost.
Repository
Built a web-first AI video generation platform with a responsive chat-based dashboard for text-to-video generation, long-video clip extraction, and multimodal guided content creation. Integrated GMI Cloud models including Kimi for prompt refinement, NVIDIA Nemotron for multimodal analysis, and PixVerse 5.6 for video generation.
TypeScript97%CSS2.9%JavaScript0.1%
Last commit 4 months ago