Open-Weight AI Models to Run Locally

DeepSeek-OCR is an OCR model for developers processing document images and PDFs on their own hardware. It also supports research into visual text compression, representing text-bearing images with compact sets of vision tokens.

Wan2.2 is a family of local AI video models for creators and developers who want to generate clips on their own hardware. Different models handle text, still images, speech and character animation.

VibeVoice is a family of voice AI models for developers and researchers building transcription and speech generation applications. It can capture who spoke and when, or produce spoken responses from incoming text.

Show 1 generates video from text prompts on your own hardware. It's aimed at researchers and developers who want to experiment with video generation while keeping the depicted scene and motion close to the prompt.

Orpheus TTS generates speech for developers building voice applications or creating custom voices. Its Llama-based models support local inference and let you control how spoken text sounds.

DeepSeek-V3 and R1 are downloadable language models for developers who want to run text generation and reasoning on their own hardware. V3 is a general language model; R1 focuses on problems that require extended reasoning.

Cellpose is local AI software for biologists and microscopy researchers who need to identify individual cells and nuclei in images. It runs on Windows, Linux and macOS, with a graphical interface for automatic segmentation, manual annotation and correcting model results.

InstantID generates personalized portraits from a single reference photo without training a model for each person. Creators and developers can retain a person's facial identity while changing the image through text prompts. Its required InsightFace face models are restricted to non-commercial research.

Stable Diffusion 3.5 is a text-to-image model family for developers building image tools and creators producing visual work on their own infrastructure. Prompts can describe the subject and composition across photographic, painted, line-art and 3D styles.

Chroma is a text-to-image model for people who want to generate images from written prompts. Its weights are available in Safetensors format.

Zonos is a local text-to-speech model for people who want to generate speech in a reference speaker's voice. It suits projects that need control over vocal delivery as well as voice matching.

LatentSync is an open-source AI lip-sync tool for video creators and researchers who want a face's mouth movements to match supplied audio. It processes video on your own GPU, with demos covering filmed faces and anime characters.

FramePack is a desktop app for creating AI video from a still image and a written motion prompt. It's for people who want to animate images on their own hardware and inspect the clip while generation is still running.

EXAONE is a family of language models from LG AI Research for developers building multilingual AI applications on their own hardware. It combines everyday language tasks with reasoning for harder problems and supports agents that call external tools.

FlagEmbedding is a Python toolkit for developers building semantic search and retrieval-augmented generation (RAG). It runs embedding models to find relevant text and rerankers to refine the order of retrieved results.

EchoMimic turns a portrait and an audio recording into a talking-head video on your own hardware. It's intended for academic research into speech animation and controlled facial movement.

TripoSR creates a 3D object from a single image on your own hardware. It's for artists, developers and researchers who want image-based reconstruction for their projects or software.

UNO is a local AI image generation framework for researchers and creators who want to preserve subjects from reference images in generated scenes. One model handles both single-subject images and compositions with several subjects.

Higgs Audio's Higgs TTS 2 is a downloadable speech model for developers building narration and conversational audio into applications. It adapts spoken delivery to the text and can generate conversations with distinct voices.

ControlNet adds visual guidance to Stable Diffusion for artists and developers who need generated images to follow a reference's layout or shape. It lets a sketch, pose or scene map guide the result alongside a text prompt.

DeepLabCut is an open-source toolbox for researchers who want to track animal movement and behavior on their own computers without attaching physical markers. It estimates the positions of visible features you choose, with support for animals, humans and objects rather than a fixed set of species or body parts.

TRELLIS is a local AI model for artists and researchers generating 3D assets from images or text. It supports several 3D representations and can edit generated assets as well as create them.

Llama is Meta's family of language models for developers, researchers and businesses building AI applications on their own hardware or servers. Downloadable weights let you choose a model for your application and available hardware.

Bark is Suno's text-to-audio model for developers and researchers experimenting with speech and other sounds. It can depart from the supplied script, which matters if you need an exact spoken reading.

Parler-TTS is a text-to-speech library for developers adding speech to applications and researchers adapting speech models. It generates audio with a voice style you describe in text.

IC-Light changes the lighting in existing images for people editing portraits, product shots, or other foreground subjects. It lets you explore different lighting conditions using a subject image as the starting point.

YuE is an AI music generation project for songwriters, musicians and developers who want to create songs with vocals and accompaniment. Its YuE2 models produce an editable composition before rendering the audio, giving you control over the music behind the recording.

MiniCPM is a family of compact language models for developers building local assistants and coding agents. It targets on-device workloads and hardware with limited computing resources.

Chatterbox is a text-to-speech model family for developers and creators who want to generate voices on their own hardware. It suits narration, voice agents and interactive media, with models for English and multilingual speech.

FastVLM is a family of local vision-language models for developers and researchers building applications that answer questions about images. It focuses on reducing the delay before an answer starts, especially with high-resolution images.

NVIDIA Cosmos provides world models for developers building robots, autonomous vehicles and industrial vision systems. It supports reasoning about physical scenes, generating training data and learning actions before real-world deployment.

SenseVoice is a local speech recognition model for developers building voice apps or analyzing recordings. Its SenseVoiceSmall model adds information about the language, emotion and surrounding sounds to speech transcriptions.

Sana is a local AI image and video generation framework for creators and developers who want to generate media on their own hardware. It includes training and inference tools for working with its diffusion models.

Qwen2.5-Coder is a family of coding language models for developers who want a local coding assistant or a model for a self-hosted code agent. The 32B instruction-tuned model supports conversational requests about code.