BRIEF-DETAILS: Text-to-image model with enhanced UNET for better eye detail and color accuracy. Features terminal-SNR-v-prediction and requires CFG Rescale configuration.
Brief-details: A 7B parameter uncensored variant of WizardLM built on Falcon-7b, designed for unrestricted text generation without built-in alignment constraints.
Brief Details: Kosmos-2 is a multimodal large language model capable of grounded vision-language tasks, supporting phrase grounding and referring expressions with detailed visual analysis.
Brief-details: RWKV-Raven-14B is a powerful RNN-based language model combining transformer-level performance with RNN efficiency, featuring 14B parameters and trained on The Pile dataset.
Brief-details: ColorizeNet is a ControlNet-based image colorization model trained on COCO dataset, capable of transforming black and white images into colorized versions using Stable Diffusion 2.1.
Brief Details: Metharme-7B: An instruction-tuned LLaMA model optimized for fiction writing and conversation, featuring specialized prompting tokens and creative text generation capabilities.
Brief Details: A unified transformer-based model for human motion analysis, supporting 3D pose estimation, action recognition, and mesh recovery with state-of-the-art performance.
Brief-details: GALPACA-30B: A scientific language model combining GALACTICA's scientific knowledge with Alpaca's instruction-following capabilities, trained on 106B tokens.
Brief-details: QteaMix is a specialized chibi/Q-style anime art model with multiple versions (Omega being latest). Optimized for creating cute, stylized characters with high stability and quality.
Brief Details: Specialized AI art model for Chinese-style game imagery. 11.6GB package with Chinese-to-tag support, optimized for 3D game art conversion and P2P image generation.
Brief Details: Wav2Lip is a state-of-the-art lip-synchronization model that accurately maps speech audio to facial movements, supporting multiple languages and identities with high precision.
Brief Details: Low-quality Bocchi the Rock character model capable of generating multi-character anime scenes. Trained on 27K images with EveryDream1, supporting 12 characters with inpainting capabilities.
Brief-details: Anime-focused text-to-image model based on animefull-latest, fine-tuned on 15,000 filtered Pixiv images. Specializes in high-quality anime character generation.
Brief Details: A state-of-the-art speech separation model using SepFormer architecture, achieving 22.4dB SI-SNRi on WSJ0-2Mix dataset for isolating mixed audio sources.
Brief-details: A comprehensive collection of SillyTavern roleplay presets optimized for various LLMs, featuring conversation, roleplay, and story modes with clear documentation
Brief-details: Multilingual 7.5B parameter language model supporting 31 languages, trained on 500B tokens with MIT license. Excels at few-shot learning tasks.
Brief Details: RoBERTa-based sentence embedding model with 355M params, trained on 1B+ sentence pairs for semantic similarity and clustering.
Brief-details: JanusFlow-1.3B is a unified multimodal AI model combining image understanding and generation through rectified flow, built on DeepSeek-LLM with 2.05B parameters.
Brief-details: A specialized LoRA model for FLUX.1-schnell focused on facial realism, trained on 200+ high-res images with 64 network dimensions and AdamW optimizer for enhanced photorealistic face generation.
Brief-details: A 2.3B parameter weather & climate foundation model trained on MERRA-2 data, capable of both atmospheric reconstruction and forecasting across 160 variables
Brief-details: A 70B parameter Llama 3.1-based model optimized for reflection and reasoning, featuring multiple GGUF quantization options from 74.98GB to 22.24GB, with integrated thought process tokens.
