Brief Details: 4x upscaling AI model for image super-resolution, based on NMKD architecture. Trained for 178k steps. Developed by gemasai for high-quality image enhancement.
Brief-details: SwiftFormer-XS is a lightweight vision transformer achieving 78.5% ImageNet accuracy with just 0.8ms latency on iPhone 14, featuring efficient additive attention mechanism
BRIEF DETAILS: GAN model that transforms photographs into Miyazaki-style anime backgrounds, based on CartoonGAN architecture. Creates soft, painterly aesthetics similar to "Howl's Moving Castle".
Brief Details: A specialized text detection model called surya_det3 by vikp, designed for the Surya platform to identify and locate text in images.
BRIEF-DETAILS: A GGUF quantized version of QwQ-32B offering multiple compression variants from 12.4GB to 34.9GB, with Q4_K variants recommended for optimal speed-quality balance.
BRIEF-DETAILS: Llama-4-Scout-17B-16E MLX format model optimized for visual instruction tasks, quantized to 6-bit precision for efficient deployment
BRIEF-DETAILS: MLX-optimized 4-bit quantized version of Llama-4-Maverick-17B, converted from meta-llama's 128E instruction model for efficient deployment on Apple Silicon
Brief-details: Meta's Llama 4 Scout model (17B parameters) optimized by Unsloth for fine-tuning, featuring multimodal capabilities and 16-expert MoE architecture.
BRIEF-DETAILS: 7B parameter LLM utilizing Token Temperature Mechanism (TTM) for enhanced reasoning, distilled from upcoming 400B model, developed by SILX AI
Brief-details: German text-to-speech model based on Orpheus-3B with synthetic voice generation, supporting 4 speakers, 12 emotions, and various outbursts for expressive speech synthesis.
BRIEF DETAILS: A 111B parameter LLM focused on creative storytelling with balanced good/evil capabilities, enhanced intelligence, and unrestricted content generation.
Brief-details: A fine-tuned vision-language model for detecting mature content in images, achieving 82.96% accuracy across 5 categories including anime, neutral, and adult content.
BRIEF DETAILS: Optimized GGUF quantized version of ChatWaifu 32B model with multiple compression variants, ranging from 7.4GB to 27GB, offering various quality-performance tradeoffs.
BRIEF-DETAILS: 27B parameter LLaMA-based model with multiple quantization options (Q2-Q8) for different hardware constraints, optimized for both performance and efficiency.
Brief Details: A Makoto Shinkai style LoRA model for FLUX.1-dev, specializing in anime-style image generation with characteristic atmospheric effects and lighting
A 32B parameter LLM optimized for software development tasks, featuring INT4 quantization and 128K context window. Achieves 37.2% resolve rate on SWE-Bench Verified, comparable to much larger models.
Brief-details: Comprehensive GGUF quantization collection of OpenHands 7B LM, offering multiple compression levels from 2.78GB to 15.24GB with varying quality-size tradeoffs.
Brief Details: Uncensored version of Qwen2.5-VL-32B-Instruct for vision-language tasks, featuring abliterated text processing while maintaining original image capabilities.
Brief-details: A specialized LoRA model for Wan2.1 14B I2V 480p that generates realistic laughing animations from static images, trained on 55 seconds of curated video clips.
BRIEF-DETAILS: Specialized LoRA trained on Wan2.1 14B I2V 480p for generating realistic kissing animations. 30 epochs training on curated clips with consistent results.
Brief Details: A specialized LoRA model for Wan2.1 14B I2V that enables creating videos of people taking selfies with their younger selves, trained on 40 seconds of curated video data.