Open-Weight AI Models to Run Locally

BrushNet fills selected areas of an image using a text prompt while retaining the surrounding content. It's for developers and researchers who want local image editing with pretrained diffusion models.

AudioLDM 2 is a local AI audio generator for sound designers and researchers experimenting with synthetic sound. It uses pretrained models to produce audio from text and provides a Gradio browser interface alongside command-line access.

Segment Anything 2 (SAM 2) is Meta's model for creating object masks in images and videos. It's for developers and researchers building visual applications or annotating footage, with interactive selection and correction.

JoyCaption generates image descriptions for people preparing datasets to train or fine-tune diffusion models. It can caption images on your own GPU, reducing the work of writing training descriptions by hand.

Ferret is a local vision-language model for researchers studying how AI understands specific parts of an image. It supports questions about selected regions and can identify image regions in its answers. Its code and data are restricted to research use.

LLaVA is a vision-language model project for researchers and developers who want to ask questions about images on their own hardware.

IndicF5 is a text-to-speech model for developers and researchers who need to generate spoken audio in Indian languages. It supports Assamese, Bengali, Gujarati, Hindi, Kannada, Malayalam, Marathi, Odia, Punjabi, Tamil and Telugu.

TorchGeo is an open source Python library for machine learning researchers and remote sensing specialists working with satellite and aerial imagery. It brings geographic datasets and pretrained models into PyTorch training workflows. The MIT license allows you to use and adapt it for your own projects.

Nomic Embed Text v1.5 is an English text embedding model for developers building search and RAG applications. It represents text numerically so applications can find passages with similar meanings.

MiniCPM-V is a family of local vision-language models for developers building applications that interpret images and video. It can run on phones and other hardware you control.

VAR is an image generation research toolkit for researchers and developers studying autoregressive models as an alternative to diffusion. It generates images through next-scale prediction, building finer detail across successive image scales.

HD-BET uses neural networks to separate brain tissue from surrounding structures in MRI scans. It runs locally and is aimed at researchers and medical imaging teams who need automated brain masks, including for scans affected by tumors, lesions or surgical resection cavities.

SmolLM3 is a compact language model from Hugging Face for developers building local AI applications and researchers studying model training. It has 3B parameters and comes in base and instruction-tuned variants.

IP-Adapter adds reference-image prompting to Stable Diffusion for artists and developers who want to guide generation with an image rather than describe every detail in words. It works with an existing text-to-image model and can combine image references with text prompts.

Seamless Communication is Meta's family of AI models for developers and researchers building multilingual speech applications. It covers translation and transcription, with dedicated models for conversations and expressive speech.

PhotoMaker generates personalized portraits and artwork from reference images and text prompts. It's for people creating photos or avatars that retain a person's recognizable features across different settings and styles.

Command A is an open weights language model from Cohere and Cohere Labs for researchers and developers building document assistants, chatbots and AI agents. It handles text input and output, with a focus on multilingual business tasks.

InsightFace is a face and person analysis toolkit for developers building identity verification, access control and media applications. Its Python tools and recognition server can process data on your own hardware.

E5 Embeddings is a family of text embedding models for developers building search and retrieval systems. It represents text as numerical vectors so applications can match queries with relevant passages on their own hardware.

Depth Anything estimates depth from ordinary images for developers and researchers who need depth maps without a depth camera. Its pretrained models show which parts of a scene are nearer or farther away; separate fine-tuned models estimate metric depth.

PaddleSeg is an image segmentation toolkit for developers building computer vision applications on their own hardware. It covers annotation, training and deployment for tasks such as medical image analysis, industrial inspection and portrait effects.

PaddleOCR is an OCR and document parsing toolkit for developers turning PDFs and images into data for document search, RAG systems and AI agents. Its local models cover both structured pages and text found in photographs.

Yi is a family of language models from 01.AI for developers and researchers who want to run English and Chinese AI on their own hardware. It includes chat models for conversation and base models you can adapt to specific tasks.

IndicTrans2 is a family of translation models for developers building multilingual applications and researchers working on Indian languages. It covers all 22 scheduled languages of India, including languages with limited training data.

InternVL is a family of vision-language models for developers and researchers building AI applications that interpret images and answer questions about them. Downloadable models let teams host inference themselves.

BiRefNet is an AI image segmentation model for developers and image workflow builders who need to separate subjects from backgrounds. It produces foreground masks for cutouts and has dedicated models for matting, where edges can include partial transparency.

Unlimited-OCR is an OCR model for developers building document extraction workflows on their own hardware or servers. It handles individual document images and multi-page inputs, including PDFs converted to page images.

MOSS-Music is a local AI model for people who need to describe, transcribe, or analyze recorded music. It works with full tracks and combines sung lyrics with musical details in its responses.

Segment Anything for Microscopy adapts Meta AI's Segment Anything to biomedical images, with models fine-tuned on microscopy data. It's for researchers who need to label cells or structures such as mitochondria in electron microscopy volumes. The software runs locally on Windows, macOS and Linux and is open source under the MIT license.

InternLM is a family of downloadable language models for developers and researchers building AI assistants or adapting models to their own applications. It includes base models for further training and chat models tuned to follow instructions.

IndexTTS is a local text-to-speech system for people creating spoken audio and developers adding speech generation to applications. It can reproduce a voice from a single reference recording.

MOSS TTS generates speech for developers and creators working on narration, dubbing and other long audio content. It supports consistent voices across long passages and can clone a speaker from reference audio without separate training for that voice.

Dia generates spoken English conversations from a written script. It's for developers and researchers who want to produce two-speaker dialogue on their own hardware, with audio references to guide the voices and delivery.

Dolphin Document Parser converts PDFs and document images into structured content for developers working on document extraction. It handles digitally created and photographed pages, including documents that mix text with tables, formulas, and code.