Projects

### Overview: Wan 2.7 API (`wan2.7`) provides a model id, pricing, parameters, and copy-paste examples for the OpenAI-compatible APIMart endpoint. It offers a unique OpenAI-compatible endpoint with no monthly subscription, allowing users to access Wan 2.7 (`wan2.7`) and over 300 image, video, and language models. ### Key features: - **Model ID and Pricing:** Accesses Wan 2.7 (`wan2.7`) with the same base URL and authorization header, enabling direct use of the model without additional setup. - **Pricing Model:** Charges per generated second, with a default tier of $0.0664 per second for 480P resolution. - **Billing Model:** The pricing is based on the generated second, with a default tier of $0.0664 per second for 480P resolution. - **Asynchronous Processing:** Supports asynchronous tasks, returning a `task_id` for tracking progress, and reattempting failed tasks. - **Image-to-Video and Video Continuation:** Allows setting a first frame and an optional last frame for Wan I2V, or providing a source video for video continuation. ### Tech Stack: - Python ### Why it Matters: - **Cost Efficiency:** Offers a unique OpenAI-compatible endpoint with no subscription, making it cost-effective for developers. - **Model Diversity:** Accesses over 300 models including image, video, and language, providing versatility in AI applications. -

Overview: Wan 2.7 Video Edit API (`wan2.7-videoedit`) is an advanced AI video editing model that offers superior visual quality, high motion coherence, and professional-grade performance. It supports various resolutions and is compatible with OpenAI's API. Key features: - Supports various resolutions (480P, 720P, 1080P) - Compatible with OpenAI's API - Offers superior visual quality and high motion coherence - Supports optional instructions and up to four reference images for target style or appearance Tech stack: Python Why it matters: This API is widely used by teams for efficient video editing, providing a cost-effective solution without the need for monthly subscriptions or pre-calculated pricing. It simplifies the process of generating high-quality video edits with minimal setup and ensures accurate billing. Repository: f60ozqblr0/wan-2-7-videoedit-api-ru Primary language: Python GitHub stars: 12 GitHub description: Wan 2.7 Video Edit API (`wan2.7-videoedit`): model id, pricing, parameters and copy-paste examples for the OpenAI-compatible APIMart endpoint.

### Overview: Wan 2.7 Image Prompts: A Comprehensive Guide to Effective Image Prompt Recipes ### Key Features: - **Model ID:** `wan2.7-image` - **Aspect Ratios:** 1:1, 3:4, 4:3, 5:7, 8:5, 16:9, 2:3, 1:1.5, 2:1, 1:2 - **Cost per Image:** $0.0216 ### Tech Stack: - Python ### Why It Matters: - **Cost Efficiency:** The Wan 2.7 Image prompts offer a cost-effective solution for image generation, with a minimum charge of $1 and billing based on actual usage. - **Model Compatibility:** Access to over 300 different image, video, and language models through a single OpenAI-compatible endpoint. - **Real-time Pricing:** The API returns the exact cost per image, ensuring transparency and accurate billing. - **Asynchronous Processing:** Supports asynchronous calls, allowing for efficient and reliable image generation without the need for upfront subscription or prepayment. ### What Separates a Usable Recipe from a Wasted Call: - **Material vs. Adjective:** Use specific materials like "frosted glass" instead of vague descriptors. - **Light Direction:** Specify light sources like "single softbox from the left" to avoid the model guessing. - **Framing:** Use precise ratios and lens language to ensure the output fits the desired layout. ### Prompt Structure Cheat

Overview: Seedance 1.0 Pro Fast Prompts offers tested image prompt recipes, including 8 examples, along with aspect ratios, model ID (`doubao-seedance-1-0-pro-fast`), and the cost per image on the APIMart route. Key features: - 8 tested image prompt recipes - Example renders and duration - Model ID: `doubao-seedance-1-0-pro-fast` - Cost per image: $0.0079 per second Tech stack: Python Why it matters: Seedance 1.0 Pro Fast provides a cost-effective, OpenAI-compatible endpoint for generating high-quality images, making it ideal for developers looking to streamline their workflow without the need for a monthly subscription.

Overview: Seedance 1.0 Pro Quality API provides access to high-fidelity video generation models on APIMart, including text-to-video and image-to-video, billed per generated second. It offers a single endpoint compatible with OpenAI, and supports 300+ models, including Seedance 1.0 Pro Quality. Key features: - Access to high-fidelity video generation models - Single endpoint compatible with OpenAI - Supports 300+ models, including Seedance 1.0 Pro Quality - Billed per generated second - Supports 480p, 720p, and 1080p resolutions - Supports prompt descriptions, duration, resolution, and image references - Asynchronous by design, tasks can be retried in a queue Tech stack: Python Why it matters: Seedance 1.0 Pro Quality API simplifies the process of integrating high-fidelity video generation models into applications, making it easier for developers to create engaging video content without the need for complex integrations or expensive subscriptions.

### Overview: This repository contains a collection of 8 tested Wan 2.7 R2V prompts, each accompanied by an example render, duration, and cost on the APIMart route, providing insights into effective image prompt recipes. ### Key Features: - **Model ID:** `wan2.7-r2v` - **Prompts:** 8 recipes, categorized into Product Ad, Cinematic, Food Ad, and more. - **Duration:** 5 seconds each. - **Cost:** $0.0664 per second, totaling $0.2656 for the set. ### Tech Stack: - Python ### Why It Matters: - **Key:** The same Base-URL and `Authorization`-Header can be used for Wan 2.7 R2V (`wan2.7-r2v`) and 300+ other models. - **Billing:** Pay-as-you-go pricing, no Abo or Prepaid-Tarif. - **Price:** Estimated cost of $0.2656 for the entire set. - **Asynchronous:** Tasks are processed asynchronously, with `task_id` and `GET /v1/tasks/{id}` for easy tracking. ### What Separates a Usable Recipe from a Wasted Call: - **Rule 1:** Name the material, not the adjective. - **Rule 2:** Set the light explicitly. - **Rule 3:** Pin the framing. ### Prompt Structure Cheat-Sheet: - **Subject:** Concrete object or person with material. - **Light

### Overview: TIGER is a unified generative model framework designed for multimodal dialogue response generation. It addresses the challenge of generating dialogue responses that are both text and images, providing a superior multimodal conversational experience. ### Key features: - TIGER is capable of handling conversations involving any combination of modalities. - It includes a textual dialogue response generator and a Text-to-Image translator. - The system is designed to be open-domain, capable of generating responses to various dialogue contexts. - Users can customize the system by choosing decoding strategies, adjusting parameters, and modifying prompt templates. ### Tech stack: Python ### Why it matters: TIGER achieves new state-of-the-art results on both automatic and human evaluations, demonstrating its effectiveness in providing a superior multimodal conversational experience.

Overview: Seedance 2.0 API is a powerful tool for generating high-quality video content using the OpenAI-compatible APIMart endpoint. It offers a single API key for accessing 300+ models, including Seedance 2.0, with flexible pricing and real-time billing. Key features: - One key, entire catalog access - $1 minimum, pay-as-you-go pricing - Real-time billing and cost reporting - Async processing with task ID polling Tech stack: Python Why it matters: Seedance 2.0 API simplifies the process of generating professional-grade video content, making it accessible and cost-effective for teams. Repository: g-v9ai/seedance-2-0-face-api-gateway Primary language: Python GitHub stars: 11 GitHub description: Seedance 2.0 API (`seedance-2.0`): model id, pricing, parameters and copy-paste examples for the OpenAI-compatible APIMart endpoint.

Overview: Seedance 1.0 Pro API is a comprehensive OpenAI-compatible API endpoint for generating high-quality video content, offering a single endpoint for multiple models, including 300+ image, video, and language models. It provides a cost-effective solution for developers without the need for monthly subscriptions or pre-paid plans, with real-time pricing updates and detailed documentation. Key features: - Single endpoint for multiple models - OpenAI-compatible API - Model id and pricing details - Real-time pricing updates - Asynchronous processing Tech stack: Python Why it matters: Seedance 1.0 Pro API simplifies the process of integrating high-quality video generation into applications, providing a cost-effective and scalable solution for developers without the need for traditional subscription models.

### Overview: XR2IND-VR is a Unity-based VR application designed for industrial training, featuring three virtual classrooms: tutorial, assembly, and troubleshooting. The app provides interactive 3D models of routers and switches, integrated with a language model assistant (LLM) via a REST API for real-time guidance. ### Key Features: - **Virtual Classrooms**: Tutorial, Assembly, and Troubleshooting rooms. - **Interactive 3D Models**: Three routers/switches with various components. - **Language Model Assistant (LLM)**: Provides real-time guidance through speech. ### Tech Stack: Unity ### Why It Matters: XR2IND-VR blends immersive VR environments with the intelligence of language models, enhancing the training experience with real-time guidance and offline manuals for accuracy. The app is designed for deployment on Meta Quest 3, making it suitable for industrial training and troubleshooting. ### GitHub Stars: 13 ### Repository: giorgosfatouros/XR2IND-VR ### Primary Language: C# ### Deployment Setup: - **Environment Preparation**: 1. Download Unity Hub. 2. Install Unity 2022.3.19f1. 3. Install Meta Quest Link. 4. Enable Quest Link on your Meta Quest 3 headset. ### Whiteboard: - Displays steps and task statuses on whiteboards. ### 3D Models: - Interactive models of routers and switches with various components. ### Speech-to-Text Component: - Uses the Hugging Face Unity API for speech-to-text functionality

Overview: This project is an unofficial re-implementation of "Antidote: Post-fine-tuning Safety Alignment for Large Language Models against Harmful Fine-tuning Attack" (ICML2025). The implementation aims to address the safety concerns in fine-tuning-as-a-service scenarios, particularly focusing on the harmful fine-tuning attack that can lead to the forgetting of safety alignment and harmful behavior in the fine-tuned model. Key features: - Implement an unofficial re-implementation of the paper's method. - Address the safety issue in fine-tuning-as-a-service scenarios. - Focus on the harmful fine-tuning attack that can break down the safety alignment of large language models. Tech stack: Python, HuggingFace Transformers, SLURM, and SLURM scripts. Why it matters: This project is crucial for ensuring the safety and reliability of large language models in real-world applications, particularly in scenarios where fine-tuning-as-a-service is prevalent. By providing a robust solution to the harmful fine-tuning attack, it helps mitigate potential risks and maintains the integrity of AI systems.

Overview: My_AI is a private AI assistant that runs entirely on your computer using Ollama, offering 12 languages support and various features like agents, document generation, image creation, voice input/output, and more. It's designed for offline use and can be accessed via mobile or VS Code extension. Key features: - 9 specialized agents and visual workflows - Word, PDF, Excel, and PowerPoint generation - Local image generation - Two-way voice input and output - VS Code extension integration - End-to-end encrypted mobile access - 12 languages support Tech stack: Python Why it matters: My_AI provides a local alternative to ChatGPT, offering comprehensive AI capabilities and language support, making it suitable for both productivity and creative tasks.

Overview: MAVERICS is a suite of test-only visual question answering datasets created from image captions, validated by manual verification, and generated from the COCO and CC3M datasets. Key features: - Test-only visual question answering datasets - Created from COCO and CC3M datasets - Generated from image captions using VQ^2A approach - Manually verified for accuracy Tech stack: Python, JSON Why it matters: MAVERICS provides a controlled environment for evaluating visual question answering models, ensuring that the questions and answers are consistent with the original image captions, thus improving the reliability and accuracy of the evaluation.

### Overview: LangSearch is a Python package designed for easily creating semantic search-based LLM applications. It leverages the power of Large Language Models (LLMs) like ChatGPT, making it highly customizable and extensible. ### Key Features: - **Automatic Data Discovery**: Crawling websites and local data to find relevant content. - **Support for 1000+ MIME Types**: Including HTML, PDF, DOCX, TXT, PNG, MP3, MP4, etc. - **Batteries Included Pipelines**: Preprocessing data with pipelines. - **Data Persistence**: Ensures efficient data synchronization. - **Embeddings**: Utilizes `text2vec-transformers` and `CLIP` models for text and images. - **RAG Methods**: Simple QA and HyDE for retrieval augmented generation. ### Tech Stack: - Python ### Why it Matters: LangSearch simplifies the process of creating semantic search-based LLM applications, making it accessible and efficient for developers working with non-public data. It integrates well with existing Python packages and tools, ensuring a smooth workflow without the need for extensive plumbing. ### Installation: ```bash pip install langsearch ``` ### Documentation: Documentation is available [here](https://langsearch.dibya.online), and code examples are in the `examples` folder. ### Features: 1. **Automatic Data Discovery**: Crawling websites and local data. 2. **Support for 1000+ MIME Types**: Including HTML, PDF, DOCX, TXT, PNG, MP3,

### Overview: FastTTS is a high-performance Test-Time Scaling (TTS) framework designed for memory-constrained edge language models. It pairs a generator LLM with a verifier (e.g., a Process Reward Model) to find high-quality reasoning traces at inference time, eliminating system-level bottlenecks and enabling edge LLMs on single consumer GPUs to match the accuracy and latency of large cloud models. ### Key Features: - **Speculative Beam Extension**: Mitigates system stragglers caused by irregular reasoning paths. - **Asymmetric Multi-Model Memory Allocation**: Partitions memory between the generator and verifier for efficient use. - **Dynamic Prefix-Aware Scheduling**: Maximizes KV cache reuse from dynamic prefix sharing. - **Separate Processes**: Enables offloading and independent GPU memory management. - **Configurable per-request search parameters**: Via `SearchConfig`. ### Tech Stack: Python ### Why it Matters: By eliminating system-level bottlenecks, FastTTS enables edge LLMs on single consumer GPUs to match the accuracy and latency of large cloud models, making it a crucial tool for accelerating test-time scaling on the edge. ### GitHub Stars: 13 ### Repository Structure: - **`FastTTS-os/`**: Contains the main FastTTS framework, including configuration, models, and utilities. - **`models/`**: Contains generator and verifier LLM wrappers, LLM subclasses, and custom schedulers. - **`search/`**: Contains the search algorithms for beam, DVTS, and best-of-N sampling. ### Installation

Overview: Qwen-Image 2.0 提供了强大的 AI 图像生成功能，支持从文本提示生成高质量图像或对现有图像进行转换。 Key features: - 支持从文本提示生成高质量图像 - 支持对现有图像进行转换 - 提供多种模型选择，包括标准和专业模型 Tech stack: Python Why it matters: Qwen-Image 2.0 是一个 OpenAI 兼容的 API 端点，提供了一个简单的 API 环境，允许用户通过一个 API Key 访问 Qwen-Image 2.0 模型，无需复杂的订阅或预付费套餐。用户只需切换 `model` 字段即可访问不同的模型，同时支持异步任务处理，无需定制集成。

Overview: Seedance 2.0 API (`seedance-2.0`) is an OpenAI-compatible endpoint offering high-quality video generation at a cost-effective rate. It supports multiple models and resolutions, with a minimum charge of $0.04 per second. Key features: - Supports 300+ models including Seedance 2.0 - Pricing is per generated second, with a minimum charge of $0.04 - Model id, pricing, and parameters are provided for easy integration - Model page for detailed information and tier breakdown - Compatible with various resolutions and aspect ratios - Async by design, allowing for batch processing and retries Tech stack: Python Why it matters: Seedance 2.0 API is ideal for teams looking to leverage ByteDance's AI capabilities without the need for a monthly subscription, offering cost-effective and professional-grade video generation.

### Overview: Kling v3 API (`kling-v3`) is a text-to-video and image-to-video model compatible with OpenAI, providing high-resolution video generation at a cost of $0.0672 per second. It supports native audio and is billed per generated second. ### Key Features: - **Model ID and Pricing:** Accessible via the `kling-v3` model ID. - **Billing Model:** Per generated second, with resolution-based pricing. - **Model Page:** Details on the exact tiers and pricing. - **Request Parameters:** Required fields include `model`, `prompt`, `duration`, `resolution`, and `size`. - **Resolution Options:** Available in 480P, 720P, and 1080P. - **Asynchronous Nature:** Tasks are batched and retried in the background, ensuring efficient execution. ### Tech Stack: - Python ### Why It Matters: - **Unified Access:** The same base URL and header for both Kling v3 and other models. - **Cost Efficiency:** Pay only for what you use, with no upfront subscription or monthly plans. - **Real-Time Pricing:** Updated pricing information is available via the API and the pricing page. - **Asynchronous Execution:** Tasks are handled in the background, ensuring smooth and efficient processing. ### Repository: - Repository: `j0ot0/kling-v3-api-es` - GitHub Stars: 13 - GitHub Description: Kling v3 API (`kling-v3`): model id, pricing, parameters,

Overview: Kling 3.0 Turbo API (`kling-3.0-turbo`) is an OpenAI-compatible endpoint providing model id, pricing, parameters, and copy-paste examples for the APIMart endpoint. Key features: - Model id and pricing information - Parameters for text-to-video and image-to-video conversions - Assynchronous processing with task tracking - Billing by generated second and resolution Tech stack: Python Why it matters: This API offers a cost-effective solution for developers looking to integrate text-to-video and image-to-video functionalities with low latency and negative prompts, without the need for a monthly subscription. GitHub stars: 13

Overview: Open Data Stack (odctl) is a command-line tool that simplifies running multiple data engineering tools on a single machine. It connects services and dependencies for Kafka, Flink, Spark, Iceberg, Trino, ClickHouse, MLflow, Feast, Airflow, and more. Key features: - Connects multiple data engineering tools into a single environment - Uses Docker Compose to start services - Supports profiles for different environments - Provides detailed documentation and CLI reference Tech stack: Python Why it matters: odctl streamlines the process of setting up complex data pipelines, making it easier for developers to manage and deploy various tools on their laptops. GitHub stars: 13 Repository: jaehyeon-kim/odctl