Brief Details: Speaker diarization model by Pyannote (v3.1) - Advanced ML solution for identifying and tracking distinct speakers in audio content, MIT-licensed.
Brief-details: A 3B parameter instruction-tuned LLM from AMD, trained on MI300X GPUs. Outperforms existing open models and matches closed-source competitors.
Brief-details: QwQ-32B-4bit is a 4-bit quantized version of the QwQ-32B model, optimized for MLX framework deployment with reduced memory footprint while maintaining performance.
Brief Details: MistralThinker-v1.1: Specialized Mistral-24B variant optimized for roleplay and storytelling, featuring DeepSeek R1 distillation and 40% RP-focused dataset.
Brief-details: A 2.1B parameter bilingual (Korean/English) instruction-tuned LLM by Kakao, optimized for compute efficiency and strong Korean performance
Brief Details: ART_v1.0 is a groundbreaking AI model for multi-layer transparent image generation, using anonymous region layouts and offering 12x faster processing with 50+ distinct layers support.
Brief-details: A 3B parameter Japanese language model optimized for instruction-following tasks, showing strong performance in both Japanese and English benchmarks with scores of 6.51 and 7.71 respectively.
BRIEF-DETAILS: Video generation model that converts Hunyuan format for ComfyUI, supporting both image-to-video and text-to-video capabilities. Native workflow compatible.
Brief Details: A 7B parameter LLM based on Qwen2.5, optimized for roleplay with MGRPO algorithm. Features 1M context, enhanced reasoning, and creative writing capabilities.
BRIEF-DETAILS: DeepHermes-3 (8B params) unifies reasoning and standard LLM responses, featuring advanced chain-of-thought capabilities and Llama-3 architecture
Brief-details: Zonos-v0.1-hybrid is a state-of-the-art multilingual text-to-speech model supporting voice cloning, emotional control, and high-quality 44kHz audio output across multiple languages.
BRIEF DETAILS: Mistral-Small-24B-Instruct-2501 is a 24B parameter instruction-tuned language model by MistralAI, built on their successful Mistral architecture.
BRIEF-DETAILS: Stability AI's medium-sized Stable Diffusion 3.5 model for image generation, featuring improved quality and performance compared to previous versions.
Brief-details: Quantized versions of Microsoft's Phi-4-mini-instruct model, offering various compression levels (Q2-Q8) for different hardware configurations and performance needs
BRIEF-DETAILS: YandexGPT-5-Lite-8B is a powerful pretrained language model with 8B parameters, 32k context length, and optimized Russian language processing capabilities.
BRIEF-DETAILS: SOTA monocular depth estimation model using knowledge distillation, offering various model sizes for efficient depth prediction from single images.
Brief-details: A state-of-the-art text-to-video model with 30B parameters, capable of generating long videos up to 204 frames with 16x16 spatial and 8x temporal compression
Brief-details: A 130B parameter multimodal LLM for audio processing, achieving state-of-the-art performance in speech recognition, understanding, and generation with superior factuality (66.4%) and chat scores (4.11).
Brief-details: Advanced hallucination detection model based on ModernBERT-large, capable of processing 8192 tokens for RAG applications. Achieves 79.22% F1 score.
Brief-details: A 3B parameter vision-language model capable of understanding images, videos up to 1+ hour, and UI interactions. Excels at visual recognition, event detection, and structured data extraction.
Brief-details: An 8B parameter LLM distilled from DeepSeek-R1, based on Llama-3.1-8B architecture. Optimized for reasoning tasks with strong math and coding capabilities.
