BRIEF-DETAILS: DiffRhythm-vae is a groundbreaking diffusion-based model for full-length song generation, combining VAE architecture with latent diffusion for fast and efficient music creation.
Brief-details: GGUF conversion of Wan2.1-I2V-14B-720P for ComfyUI - 14B parameter image-to-video model optimized for 720P resolution with multiple quantization options
Brief Details: Open-weight TTS model with 82M parameters supporting English & Chinese. Features 100+ Chinese speakers and 3 synthetic English voices. Built on StyleTTS 2 architecture.
BRIEF DETAILS: Wan2.1-I2V-14B-480P is a 14B parameter image-to-video generation model capable of producing high-quality 480P videos, featuring efficient processing and SOTA performance.
Brief-details: 8B parameter multilingual LLM from IBM with enhanced reasoning capabilities, supporting 12 languages and specialized for instruction-following tasks with controllable thinking ability.
Brief Details: HunyuanVideo is an open-source video foundation model with 13B parameters, capable of high-quality text-to-video and image-to-video generation using advanced 3D VAE and MLLM architecture.
Brief Details: Smart-turn is an AI model by pipecat-ai available on HuggingFace, designed for advanced text processing and transformation capabilities.
🤖 Brief Details: Advanced 3D asset generation model from Tencent featuring two-stage pipeline for mesh creation and texturing, achieving SOTA performance in 3D generation with CLIP score of 0.809.
Brief-details: A sophisticated anime image tagging model capable of identifying 70,527 tags across 7 categories, achieving 61% F1 score. Trained on consumer hardware with innovative two-stage architecture.
Brief-details: DeepSeek-R1-GGUF is a quantized version of DeepSeek-R1, a powerful reasoning model with 671B parameters (37B activated), optimized for mathematical and logical tasks.
BRIEF DETAILS: Uncensored variant of DeepSeek-671B using abliteration technique. Aims to remove content restrictions while maintaining core capabilities. Part of huihui-ai's abliterated model series.
Brief-details: A powerful 32B parameter math-focused model achieving SOTA AIME24 scores (76.6%). Trained via curriculum SFT & DPO for only $1000, surpassing DeepSeek-R1.
Brief-details: Optimized ComfyUI-compatible version of HunyuanVideo for AI video generation, featuring safetensors and fp8 formats with FastVideo's distilled version support
Brief-details: Mistral-7B-Instruct-v0.3 is a powerful 7B parameter instruction-tuned language model from MistralAI, optimized for following complex instructions and generating high-quality responses.
Brief Details: A highly optimized Stable Diffusion 3.5 Large variant delivering 6x faster image generation with superior quality and compatibility, featuring 8-step inference and specialized scheduling
Brief-details: GGUF conversion of Wan2.1-T2V-14B model for text-to-video generation, optimized for ComfyUI integration with FP16 quantization support.
BRIEF-DETAILS: 32B parameter distilled model from DeepSeek-R1, achieving SOTA performance in math and reasoning tasks, outperforming OpenAI-o1-mini across benchmarks
Brief-details: Powerful hybrid SSM-Transformer model with 94B active parameters, 256K context length, superior long-context handling and 2.5X faster inference than comparable models.
Brief-details: Meta's 8B parameter instruction-tuned LLaMA model optimized for dialogue and instruction following, part of the LLaMA 3.x series
Brief-details: A 70B parameter decensored LLaMA 3.3 variant built on Deepseek's R1 Distill, designed for unrestricted creative expression and adversarial responses.
Brief-details: GGUF conversion of Wan2.1-I2V-14B-480P model optimized for 480p video generation, featuring 14B parameters and multiple quantization options for ComfyUI integration
