Open-Weight AI Models to Run Locally

Chatterbox is an MIT-licensed text-to-speech model family for developers and creators who want to generate speech on their own hardware. You can self-host it on a GPU, including in an air-gapped environment, without an account or API key. Resemble AI also offers separate managed hosting.

Gemma is Google DeepMind’s family of open-weight AI models for developers building applications that can run on their own hardware. Its range covers compact models for phones and IoT devices alongside larger Gemma 4 models for reasoning on personal computers and servers. Some applications can work offline, keeping model inference on the device. Google AI Studio and Google Cloud are also available for hosted use.

IndexTTS, currently IndexTTS-2.5, is a local text-to-speech system that can reproduce a speaker's voice using one reference recording. It's for people creating spoken audio and developers building speech generation into their own applications. Voice identity and emotion have separate controls, so an emotional reference can shape the delivery while a different recording supplies the voice.

Wan2.2 is an open source family of video generation models for creators and developers who want to make clips on their own GPUs. Licensed under Apache 2.0, it covers text-to-video and image-to-video generation, with separate models for speech-driven video and character animation. Generation runs on your hardware when you use the downloadable models.

Fish Speech, currently featuring Fish Audio S2 Pro, is a self-hosted text-to-speech system for creators producing narration and developers building voice applications. It combines voice cloning with control over emotion and delivery within a script. Code and model weights use the custom FISH AUDIO RESEARCH LICENSE.

Qwen3 is a family of language models from Alibaba Cloud’s Qwen team for people who want to run models locally or on their own servers. It spans smaller and larger dense models as well as mixture-of-experts models. The weights are publicly available.

VibeVoice is a family of MIT-licensed, open-source voice AI models for developers and researchers building local transcription or speech generation tools. Its speech recognition models combine transcript text with speaker labels and timestamps, so recordings retain information about who spoke and when.

Demucs separates a finished song into vocals, drums, bass and the remaining accompaniment on your own computer. It's for musicians who need individual stems or a vocal-free backing track, and developers building audio tools. The project is archived and no longer maintained. Its Python code is open source under the MIT license.

Qwen3-Embedding is a family of text embedding models for developers building search and document analysis on their own hardware or servers. It turns text into numerical representations that applications can compare by meaning. Its multilingual support covers languages including English, Chinese, Arabic and Ukrainian, as well as programming languages.

FlagEmbedding is an open-source Python toolkit for developers building semantic search or retrieval-augmented generation (RAG) into their own applications. It runs BGE embedding and reranking models, with tools to fine-tune both and evaluate retrieval results. The library uses the MIT license.

Llama Guard is Meta's collection of downloadable AI content moderation models for developers building LLM applications. It checks user inputs and model responses for content that violates safety policies, including text and images. Developers can use the models in their own deployments or access moderation through Meta's hosted Llama API.

AudioLDM 2 generates sound effects, music and speech on your own hardware. It's a Python tool for people experimenting with synthetic audio, including sound designers and researchers who want to work with pretrained models. A Gradio browser interface and command-line tools provide access to local generation; a hosted Hugging Face demo is also available.

TripoSR reconstructs a 3D object from a single image on your own hardware. Developed by Tripo AI and Stability AI, it's an open-source model for researchers, developers, and artists who want image-based 3D generation they can use in their own projects. The MIT license covers both the source code and pretrained models.

Nomic Embed Text v1.5 is an English text embedding model for developers building semantic search, document retrieval, and RAG applications on their own hardware or servers. It turns text into numerical representations that applications can compare by meaning. Its main distinction is adjustable embedding size: you can use smaller vectors when storage matters, with a tradeoff in retrieval quality.

DiffRhythm is a local AI music generation model for musicians, developers and researchers who want to create full-length songs on their own hardware. It uses latent diffusion to generate songs with vocals and accompaniment, and can also produce instrumental music. The full model supports songs up to 4 minutes and 45 seconds.

Stable Audio Open is a text-to-audio model you can run on your own hardware to generate sound effects, field recordings and music samples. It's aimed at artists and machine learning practitioners experimenting with audio generation, and it performs better on environmental sounds and effects than on music.

EmbeddingGemma is a text embedding model for developers building search and document features that run on phones, laptops or tablets. Based on Gemma 3, it converts text into numerical representations so applications can find related passages by meaning. Embeddings stay on your hardware, and the model works without an internet connection.

Snowflake Arctic Embed is a family of open-source text embedding models for developers building semantic search and document retrieval systems. It turns queries and documents into numerical representations that a search system can compare by meaning. The models use the Apache 2.0 license.

GTE (General Text Embedding) is Alibaba’s family of downloadable models for representing text as vectors. Developers use these representations to compare queries with documents, cluster related text or supply retrieval components for larger applications.

mxbai-embed is Mixedbread’s downloadable text embedding model family for developers building their own retrieval systems. It converts queries and passages into vectors that an application can compare for relevance or similarity.

ModernBERT is a family of open-source text encoder models for developers building document search, classification and code retrieval on their own hardware. Its longer context lets it process documents and code passages that exceed the limits of older BERT models. The code and models use the Apache 2.0 license.

ColPali is a local AI document retrieval library for developers and researchers building document search or retrieval-augmented generation systems. It searches pages as images, using their text, charts and layout together rather than relying on a separate OCR pipeline. The colpali-engine package is deprecated; its maintainers recommend Sentence Transformers for new projects and production use.

Jina Embeddings is a family of models that converts content into vectors for retrieval, similarity matching, classification and clustering. It includes multilingual text models and multimodal variants for searching across different media.

Higgs Audio is a family of text-to-speech models from Boson AI for developers building narration and conversational audio. Higgs TTS 2 can adapt pacing and intonation to the text and generate dialogue with distinct speakers across multiple languages.

E5 Embeddings is a family of text embedding models for developers building search and retrieval systems on their own hardware. It converts text into numerical representations for matching queries with relevant passages. The family includes English and multilingual models, plus instruction-based variants for task-specific embeddings.

WhisperSpeech is a local text-to-speech system that uses OpenAI Whisper as the basis for generating speech. It's for developers and speech researchers who want to work with downloadable models on their own hardware. It supports voice cloning.

Zonos is an open-source text-to-speech model for people who want to generate speech and clone voices on their own hardware. It can match a speaker from a short reference recording, with controls for delivery and emotion. The code uses the Apache 2.0 license.

StyleTTS 2 is an open-source text-to-speech model for developers and speech researchers who want to generate expressive speech on their own hardware. It can choose a speaking style from the text without a reference recording, while its multispeaker model uses reference audio to reproduce a speaker's voice and delivery. The Python code uses PyTorch and carries the MIT license.

Sesame CSM is an open-source speech generation model for developers and researchers building voice applications on their own hardware. It uses text and audio inputs to generate speech, with support for conversational context and different speakers. It's a model component for applications that need spoken output.

Pyramid Flow is an open-source AI video generator for people who want to create clips on their own hardware, and for researchers working on video models. It turns text prompts into video or animates a supplied image with guidance from text. The Python code uses PyTorch and carries MIT. Model licenses are separate: the SD3-derived weights use the Stability AI Community License. Check the chosen checkpoint before deployment.

Voxtral is Mistral AI's open-source audio and text model for developers building self-hosted speech applications. It can answer questions about recordings and produce structured summaries within the same model that transcribes speech. Mistral's separate mistral-inference library is archived and no longer maintained; Voxtral supports vLLM and Hugging Face Transformers.

XTTS v2 generates speech from text using a reference voice recording or a preset speaker. It runs locally through Coqui TTS and suits developers building speech into apps, as well as researchers who want to fine-tune a speech model on their own hardware.

AnimateDiff adds text-driven animation to personalized Stable Diffusion models without requiring separate training for each model. It's for artists and developers who want to generate motion while keeping the visual style of a chosen image model. The Python implementation runs locally and includes a Gradio browser interface.

Stable Video Diffusion is an image-to-video model from Stability AI that animates a supplied still image into a short clip. It is for developers and researchers working with video generation on their own hardware. The SVD model generates 14 frames; SVD-XT is fine-tuned for 25 frames at 576×1024 resolution.

Open-Sora is an open source AI video generation project for developers, researchers, and creators who want to run and adapt a model on their own hardware. Its model focuses on turning reference images into video, with text prompts guiding the result. It also generates video directly from text. The code and Open-Sora 2.0 weights use Apache 2.0.

SkyReels-V2 is a family of downloadable video models for developers and researchers who want to run generation on their own GPUs. It supports text-to-video, image-to-video and extended video generation. Its autoregressive Diffusion Forcing architecture generates longer sequences by continuing video segments.