gradio

Official PyTorch implementation of AFP-GIC: controllable generative image compression with a single model for multiple bitrates.

端到端音乐情感分析：37 维声学特征工程 → 传统机器学习分类 → Gradio 交互演示。基于 GTZAN，含弱监督局限声明。

Verified Al Rajhi bank-statement reader (Arabic-first): literal VLM transcription → balance chain + printed-footer oracle → Arabic RAG Q&A. Sample scope; synthetic public fixture only.

Analyze suspicious messages and screenshots from Pakistan to identify fraud and verify notices before taking action.

Detect deepfake images and videos with EfficientNet-B0, face extraction, Grad-CAM heatmaps, and a FastAPI REST API for clear analysis

🎙️ Implement fast, dependency-free C inference for Qwen3-ASR speech-to-text models with efficient streaming on modest hardware.

🎤 Run Qwen3-TTS locally on Apple Silicon for offline AI text-to-speech, offering voice cloning and customization features without cloud reliance.

🎥 Enable real-time image captioning and classification with this Jupyter notebook collection, featuring pretrained models and live webcam applications.

🖼️ Edit images swiftly with the Qwen-Image-Edit-2511 model, featuring lazy-loaded LoRAs and interactive 3D visualization for seamless single-image transformations.

🖼️ Streamline GUI interaction by converting UI screenshots and task instructions into structured tool calls with visual feedback and precise actions.

🖼️ Enhance image text recognition with HunyuanOCR-Demo, a Gradio app for precise OCR tasks, including text detection, extraction, and formatting.

🚀 Deploy InfiniteTalk and MultiTalk models effortlessly with automated management, Docker support, and a user-friendly Gradio interface.

🖼️ Enhance image and text understanding with Qwen-3VL, a web app for multimodal tasks like captioning and object detection using state-of-the-art AI.

🏡 Predict real estate prices in Madrid using a machine learning model, with an interactive web app to visualize price estimates for potential properties.

🚀 Optimize Hugging Face models on macOS with this starter kit, featuring Jupyter, Accelerate, and Gradio for streamlined development.

RAG guide for F-1 OPT, CPT and STEM OPT questions. Retrieves from government, university and attorney sources in ChromaDB (all-MiniLM-L6-v2), answers with Groq llama-3.3-70b, and lists the sources used. Built for CodePath AI201, 2026.

Fine-tuned DistilBERT classifier that labels Hacker News comments as analysis, assertion, or reaction. A zero-shot Llama-3.3-70B baseline won, 0.781 to 0.625 accuracy (test n=32). Built for CodePath AI201, 2026.

A grid/gallery-based style selector extension for all major Stable Diffusion WebUI builds.