Projects

### Overview: Seedance 2.0 API (`seedance-2.0`) is an OpenAI-compatible endpoint that provides high-quality video generation services. It supports text-to-video and image-to-video models, offering cinematic motion and HD output at a cost of $0.066 per second for 480P resolution. ### Key Features: - **Model ID and Routes:** Access to Seedance 2.0 (`seedance-2.0`) via the `seedance-2.0` model ID. - **Pricing:** Per generated second, with a default resolution of 480P, $0.04 for 720P, and $0.216 for 1080P. - **Billing Model:** Uniquely priced based on the number of generated seconds, with a minimum charge of $1 per task. - **Asynchronous Processing:** Tasks are completed and reported in real-time, with the cost of each task included in the response. ### Tech Stack: - Python ### Why It Matters: - **One Key, Multiple Models:** The same base URL and headers provide access to Seedance 2.0 (`seedance-2.0`) and over 300 other image, video, and language models. - **Cost-Effective:** No need for pre-paid plans or monthly subscriptions; charges are applied based on usage. - **Real-Time Pricing:** The pricing is updated in real-time, and tasks report the exact cost in the `cost` field. - **As

Overview: Seedance 1.0 Pro API is a fast, OpenAI-compatible model endpoint for generating high-quality video content using the APIMart service. It provides model id, pricing, and parameters for easy integration and usage. Key features: - Model id for easy identification - Pricing based on generated seconds, with tiered billing - Parameters for prompt description, duration, resolution, and framing - Integration with APIMart for easy access and billing Tech stack: Python Why it matters: Seedance 1.0 Pro API simplifies the process of integrating high-quality video generation into projects, providing a fast and efficient solution for video production and content creation. Repository: jd6ymi6bby8gzp/seedance-1-0-pro-api-ru Primary language: Python GitHub stars: 13 GitHub description: Seedance 1.0 Pro API (`seedance-1.0-pro`): model id, pricing, parameters and copy-paste examples for the OpenAI-compatible APIMart endpoint.

Overview: This toolkit allows for the incorporation of multimodal data, including images and tabular data, with HuggingFace Transformers for classification and regression tasks. Key features: - Integrates HuggingFace Transformers for text features - Adds a combining module that combines outputs from the transformer with categorical and numerical features - Supports multiple pretrained transformers for tabular data, including BERT, ALBERT, DistilBERT, RoBERTa, XLM, XLNET, and XLM-RoBERTa - Includes datasets for text and tabular data, such as Women's Clothing E-Commerce Reviews, Melbourne Airbnb Open Data, and PetFindermy Adoption Prediction Tech stack: Python, PyTorch, transformers 3.1 Why it matters: This toolkit provides a comprehensive solution for incorporating multimodal data into text-based models, enhancing their capabilities for tasks like recommendation prediction, price estimation, and pet adoption speed prediction.

Overview: Agnes AI Creator 是基于免费 Agnes AI 的自托管多模态 Web 客户端，支持 AI 对话、文生图/图生图、文生视频/图生视频。它使用 Vue 3 + FastAPI 全栈技术，提供毛玻璃现代 UI 和七牛云持久化功能。 Key features: - AI 对话 - 文生图/图生图 - 文生视频/图生视频 - 毛玻璃现代 UI - Vue 3 + FastAPI 全栈 - 七牛云持久化 Tech stack: Vue 3 + FastAPI Why it matters: Agnes AI Creator 是一个免费使用 Agnes AI 模型的自托管多模态 Web 客户端，提供流式对话、图片生成和视频生成等功能，适用于需要快速生成内容的场景。

### Overview: This project is a Python-based MCP server designed to read WeChat Official Account articles, returning images and video keyframes as content blocks rather than URLs. This approach allows for a more seamless and efficient integration with large language models. ### Key Features: - **Reliable WeChat Scrape**: Pure Python `httpx` GET, no need for Rust binary or headless browser. - **Native Image Content**: PNG/JPG images returned as MCP `Image` blocks, with GIFs filtered and capped at 10 per article. - **Video Handling**: Supports WeChat Official Account native videos, Tencent Video keyframes, and WeChat Channels video keyframes. - **Publish Time Recovery**: Extracts the `var ct` Unix timestamp that other parsers miss. - **Minimal Install**: Requires `pip install weixin-articles-mcp` and optional `ffmpeg` for video processing. ### Tech Stack: Python 3.10+ with `httpx` for HTTP requests and `ffmpeg` for video processing. ### Why It Matters: This tool provides a more efficient and seamless integration with large language models, allowing them to see the actual images and video content directly, rather than having to click through to each image or video separately. ### Repository: jj-cheng25/weixin-articles-mcp ### GitHub Stars: 13 ### GitHub Description: MCP server for reading WeChat Official Account articles with native multimodal output — images and video keyframes as content blocks, not URLs. For personal/research use. This tool reads only publicly accessible article URLs

Overview: Grok Imagine Image 2.0 API is a high-fidelity image generation service that supports OpenAI-compatible endpoints, offering pricing based on image resolution and quality. Key features: - Supports multiple image, video, and language models - Provides a single endpoint compatible with OpenAI - Offers pricing per image by resolution and quality - Supports both low and medium resolution tiers - Allows for flexible image generation with customizable parameters Tech stack: Python Why it matters: Grok Imagine Image 2.0 API simplifies image generation by providing a single endpoint that integrates with OpenAI's API, making it an efficient and cost-effective solution for developers looking to enhance their applications with high-fidelity images.

Overview: Wan 2.7 Prompts is a repository of 8 tested recipes for generating high-quality images using the Wan 2.7 model, including detailed prompts and pricing information. Key features: - 8 tested recipes with example renders - Model: Wan 2.7 - Effective price: $0.0664 per second - Estimated cost for the entire set: $0.2656 Tech stack: Python Why it matters: This repository provides developers with practical, tested recipes for generating high-quality images using the Wan 2.7 model, ensuring consistent and effective results while keeping costs under control.

Overview: Seedream 4.5 API (`seedream-4-5`) provides a cost-effective and OpenAI-compatible endpoint for generating high-quality images and videos. With a base price of $0.026 per image, it allows for flexible pricing based on usage, making it ideal for developers looking to integrate image generation into their projects without the need for a subscription. Key features: - **Cost-Effective Pricing**: $0.026 per image, with no monthly subscription or upfront costs. - **OpenAI-Compatible Endpoint**: Compatible with OpenAI's API, ensuring seamless integration with existing AI models. - **Flexible Pricing Model**: Based on the number of images generated, offering a tiered pricing structure for different resolutions. - **Asynchronous Processing**: Tasks are processed asynchronously, allowing for real-time updates and easy monitoring with `GET /v1/tasks/{task_id}`. Tech stack: Python Why it matters: Seedream 4.5 API is a cost-effective solution for developers looking to integrate image generation into their projects without the need for a subscription. Its OpenAI-compatible endpoint ensures compatibility with existing AI models, making it a valuable addition to any developer's toolkit. Repository: jnh7w0hfhm7dr/seedream-4-5-api-id Primary language: Python GitHub stars: 13 GitHub description: Seedream 4.5 API (`seedream-4-5`): model id, pricing, parameters and copy-paste examples for the OpenAI-compatible APIMart endpoint.

### Overview: Nano Banana 2 API (`gemini-3.1-flash-image-preview`) is an OpenAI-compatible endpoint that provides fast AI image generation with native 4K output and editing capabilities. It offers a scalable and cost-effective solution for developers, allowing one API key to access over 300+ image models. ### Key Features: - **One Key, All Catalogs:** A single API key grants access to Nano Banana 2 and 300+ other image models, using the same base URL and authorization header. - **Minimal Cost:** The pricing is per delivered image, with a default cost of $0.015 per image. The effective price varies based on the output resolution. - **Summation of Costs:** Each completed task reports the exact cost in the `cost` field. - **Asynchronous Processing:** Tasks are processed asynchronously, so you can monitor progress by polling the task ID. ### Tech Stack: - Python ### Why It Matters: - **One Key, All Models:** The same base URL and authorization header provide access to Nano Banana 2 and over 300 other image models, reducing the need for multiple keys. - **Minimal Cost:** The effective cost is $0.015 per image, making it a cost-effective solution. - **Summation of Costs:** Each task reports the exact cost, ensuring accurate billing. - **Asynchronous Processing:** Tasks are processed asynchronously, allowing for efficient monitoring and management. ### GitHub Stars: 13 ### Repository: - **Repository:** jnp3

### Overview: This repository contains a tuning recipe for DeepSeek-V4-Flash-0731 on a 4x DGX Spark cluster, focusing on a specific workload generated by Hermes. The recipe includes detailed measurements and values that were lost during the optimization process, providing insights into the performance and trade-offs of the chosen configuration. ### Key Features: - **Tuning DeepSeek-V4-Flash-0731 on 4x DGX Spark (GB10) at TP=4** - **Measurements and Values Lost:** - `draft_sample_method`: Greedy vs Probabilistic (26.5% vs 34.3% acceptance) - `num_speculative_tokens`: 7 vs 115.2 tok/s, 102.6 tok/s (k=8 vs k=10) - `max_cudagraph_capture_size`: 96 vs 36 (truncates to 32, drops to eager above 4 seqs) - `gpu_memory_utilization`: 0.85 vs 0.80 (−6.9 GiB, drops to eager above 4 seqs) - `max_num_batched_tokens`: 8264 vs 8192 (lands under vLLM's own warning threshold) - `max_model_len`: 393,216 vs 327K…1M (KV pool flat the whole way) ### Tech Stack: - Python

Overview: LogosDB is a fast, memory-efficient semantic vector database optimized for AI agents and RAG systems. It provides approximate nearest-neighbor search using HNSW, supports text metadata, and integrates with LangChain and LlamaIndex. Key features: - Flat binary storage for vectors and metadata - Approximate nearest-neighbor search with O(log n) query time - Optional text and timestamp metadata - Timestamp range filtering and multiple distance metrics - Bulk vector access for GPU loading - Thread-safe API - Crash recovery with HNSW index backfill - Scales to millions of vectors - Framework integrations with LangChain and LlamaIndex - MCP server integration via Claude Code Tech stack: C++ Why it matters: LogosDB is specifically designed for embedding-based memory retrieval in LLM inference, making it highly optimized for AI agent and RAG systems. Repository: jose-compu/logosdb GitHub stars: 13

### Overview: Gemini Nano Banana 2.1 Prompts: Tested Image Prompt Recipes with Example Renders, Aspect Ratios, Model ID, and Cost Per Image ### Key Features: - **Model ID:** `gemini-nano-banana-2.1` - **Aspect Ratios:** 1:1, 4:3, 8:5, etc. - **Cost Per Image:** $0.0137 - **Example Renders:** Product hero shot, food photography, and more ### Tech Stack: Python ### Why It Matters: - **One Key — All Catalog.** Access Gemini Nano Banana 2.1 (`gemini-nano-banana-2.1`) and 300+ other models with the same base URL and header. - **Minimal Cost.** No need for monthly billing; pay as you go with $1 per image. - **Returnable Charges.** The cost is returned in the response, so you can track your progress. - **Asynchronous Processing.** Send tasks and get `task_id`; check `GET /v1/tasks/{id}` for results. ### What Separates a Usable Recipe from a Wasted Call: - **Material vs. Adjective:** "Frosted glass" vs. "beautiful" or "high quality". - **Light Direction:** "Single softbox from the left" vs. "golden hour backlight". - **Framing:** "3:4, medium format look" vs. "macro, overhead flat lay". ###

Overview: Wan 2.7 API (`wan2.7`), an OpenAI-compatible endpoint, provides model id, pricing, and parameters for image-to-video tasks on APIMart. Key features: - Supports Wan 2.7 (`wan2.7`) and over 300 other image, video, and language models. - Pricing is per generated second, billed by resolution. - Supports first frame, last frame, and source video continuation. - Asynchronous by design, tasks can be retried in a queue. Tech stack: Python. Why it matters: Simplifies integration with OpenAI-compatible models, offering superior visual quality and performance at a cost-effective rate. Repository: jtyth8d82sy1wf/wan-2-7-api-fr Primary language: Python GitHub stars: 13 GitHub description: Wan 2.7 API (`wan2.7`): model id, pricing, parameters and copy-paste examples for the OpenAI-compatible APIMart endpoint.

Overview: MAI-Image-2.6 API is a model id, pricing, and parameters for the OpenAI-compatible APIMart endpoint, offering scalable image generation at a cost-effective rate. Key features: - Model id: MAI-Image-2.6 - Pricing: $0.0311 per image - Parameters: - Model: The model id shown in the route table - Prompt: What to generate or what to change when editing - Size: Aspect ratio or "auto" for route support - Resolution: Output resolution tier - N: Number of images per task (limited by route) - Image URLs: Reference images as public HTTP(S) URLs or base64 data URLs - NSFW check: Runs `omni-moderation-latest` before submission Tech stack: Python Why it matters: MAI-Image-2.6 API provides a scalable image generation service at a cost-effective rate, making it a popular choice for developers looking to integrate image generation into their applications without the need for a monthly subscription. GitHub stars: 13

### Overview: Flow Music Prompts: A collection of 8 tested recipes for generating high-quality image prompts using the Flow Music model. These recipes are designed to provide clear and effective instructions for generating professional-looking images at a low cost. ### Key Features: - **Model ID**: `flowmusic` - **Cost per Image**: $0.01 per second - **Example Renders**: 4 recipes with detailed prompts and examples - **Aspect Ratios**: 85mm, 24mm, macro, overhead flat lay - **Pricing**: Estimated cost of $0.04 for the entire set ### Tech Stack: Python ### Why It Matters: - **Cost Efficiency**: Generates high-quality images at a low cost, making it ideal for creative projects on a budget. - **Model Compatibility**: Compatible with OpenAI models, allowing seamless integration with other tools and platforms. - **Pricing Transparency**: Real-time billing and clear pricing information, ensuring cost transparency and ease of use. ### Repository: - **Repository**: `jyp5o6wi5bbg1/flow-music-prompts-id` - **Stars**: 13 - **Description**: A collection of 8 tested recipes for generating image prompts using the Flow Music model. This repository includes detailed prompts, examples, and pricing information.

### Overview: 通义万相 — Wan 3.0 API (`wan3.0-video`): 通义万相提供了一站式的视频模型访问，包括文本到视频、图像到视频和文件/网页生成。只需一个 API 密钥，即可访问 Wan 3.0（`wan3.0-video`）以及其他 300 多个模型，无需预付费套餐。最低充值 $1，按量付费，费用随响应一并返回。 ### Key features: - **一个 API 密钥，整个目录访问**：同一个 `https://api.apimart.ai/v1` 基础 URL 和 `Authorization` 请求头即可访问 Wan 3.0（`wan3.0-video`）以及图像、视频、语言等 300+ 模型。 - **$1 起充，按量付费**：没有订阅、也没有需要预先估算用量的预付费套餐：最低充值 $1，调用多少扣多少。 - **费用随响应一并返回**：每个完成的任务都会报告实际计费金额（`cost` / `credits_cost`）；状态为 `failed` 的任务会自动退款。 - **原生异步设计**：提交任务、拿到 `task_id`、轮询 `GET /v1/tasks/{id}`——批处理和重试都是普通的队列工作，不需要定制集成。 ### Tech stack: Python ### Why it matters: - **一个 Key，整个目录**：同一个

### Overview: Seedance 2.0 Fast is a faster version of ByteDance's Seedance 2.0 route on APIMart, providing text-to-video and image-to-video functionalities. It offers a per-generated-second pricing model, allowing direct access to the model through a single API key without the need for a subscription. ### Key Features: - **Direct Access:** Utilizes a single `Authorization` header for accessing the model through the APIMart endpoint. - **Pricing Model:** Per-generated-second pricing, billed by resolution. - **Model Compatibility:** Compatible with OpenAI-compatible APIMart endpoint, providing a wide range of models and parameters. - **Model Page:** Accessible through the model page provided in the README. ### Tech Stack: - Python ### Why It Matters: - **Cost Efficiency:** The per-generated-second pricing model makes it highly cost-effective, eliminating the need for a subscription. - **Model Flexibility:** Direct access to the model through a single API key allows for easy integration and experimentation. - **Model Diversity:** Access to over 300+ models, including text-to-video and image-to-video functionalities, providing versatility in content creation. ### Repository: - Repository: `k9fc4/seedance-2-0-fast-api-ko` - GitHub Stars: 13 - GitHub Description: Seedance 2.0 Fast (`doubao-seedance-2.0-fast`): model id, pricing, parameters, and copy-paste examples for the OpenAI-compatible APIMart endpoint.

Overview: RAG-Gemini is an implementation of the Retrieval-Augmented Generation model in TypeScript, utilizing the Gemini LLM. Key features: - Supports TypeScript development - Integrates with Gemini LLM for advanced language generation - Provides enhanced text generation capabilities through retrieval-based augmentation Tech stack: TypeScript Why it matters: This project showcases the integration of cutting-edge language models with modern programming languages, offering developers a powerful tool for text generation tasks. Repository: kanugurajesh/RAG-Gemini GitHub stars: 13

Overview: This repository contains tested image prompt recipes for Seedream 5.0 Lite, including 8 recipes with example renders, aspect ratios, the model ID (`seedream-5-0-lite`), and the cost per image. Key features: - 8 Seedream 5.0 Lite prompts - Example renders, aspect ratios, and model ID - Cost per image Tech stack: Python Why it matters: This repository provides practical image prompt recipes for Seedream 5.0 Lite, offering a straightforward way to generate high-quality images without the need for complex integration or prepayment plans.

### Overview: This repository contains tested Veo 3.1 prompts with example renders, aspect ratios, the model id (`veo3.1-fast-ext`), and the cost per image. The prompts are designed to help developers create high-quality images efficiently. ### Key features: - **Model ID:** `veo3.1-fast-ext` - **Render Examples:** 8 different recipes with their corresponding durations and costs. - **Cost per Image:** $0.064 per second, with a total estimated cost of $0.256 for the set. ### Tech Stack: Python ### Why it Matters: The repository provides a practical guide for developers to create effective Veo 3.1 prompts, ensuring high-quality images without the need for expensive subscription models.