Open-Weight AI Models to Run Locally

ACE-Step is a local AI music model for musicians and producers who want to generate songs or revise existing audio. It accepts text descriptions and lyrics, with control over the length of the result.

StyleGAN3 is NVIDIA's local AI image generation and training software for researchers and developers. Its architecture addresses generated detail that stays fixed to pixel positions instead of following the depicted objects as they move.

NeuTTS is a collection of on-device text-to-speech models for developers building voice assistants and speech-enabled apps. It generates audio on your own hardware rather than sending speech generation to a cloud API.

Falcon-H1 is a family of language models from the Technology Innovation Institute for developers building AI applications on their own hardware. Smaller models suit edge devices, while larger variants target server deployments.

HunyuanWorld generates explorable 3D environments from text prompts or images. It's for game developers, VR creators and researchers who need scene geometry for interactive content or physical simulation.

VideoCrafter generates video on your own hardware from written prompts or still images. It's aimed at researchers and people exploring video generation for personal use; the toolbox limits use to personal, research and non-commercial purposes.

Stable Audio Open generates audio from text for artists and researchers experimenting with sound effects, field recordings and music samples. It performs better on effects and environmental sounds than music, and it can't generate realistic vocals.

HunyuanDiT is a local text-to-image model for people generating images from Chinese or English prompts. It handles detailed descriptions and Chinese cultural references, including poetry, idioms, food and architecture.

ColBERT is a self-hosted neural search library for developers building document search or retrieval for AI applications. It finds relevant passages in your own text collections and stores their search representations on disk. The Python project is open source under the MIT license.

HunyuanCustom generates videos of specific people or objects for creators and researchers who need subjects to stay recognizable across frames. It builds on HunyuanVideo and supports generation on your own hardware.

MOSS is a conversational language model from Fudan University for developers and researchers who want to run and adapt a local LLM. It includes model weights, dialogue datasets and fine-tuning code.

Whisper is an open source speech recognition model for transcribing recordings on your own hardware. It's for developers building voice features and people working with multilingual audio.

Qwen3 is a language model family from Alibaba Cloud's Qwen team for people building chat applications or running models on their own hardware. Its public weights include dense and mixture-of-experts models in smaller and larger sizes.

AnimateDiff generates animations with personalized Stable Diffusion models for artists and developers who want motion in their chosen visual style. Compatible image models don't need separate animation training.

Qwen3-TTS is a family of local text-to-speech models for developers building speech applications or creating audio with custom voices.

Qwen2.5-VL is a vision-language model for developers building visual assistants, document processing applications and screen-based agents. It answers questions about images and video and can return structured data for software to use.

OLMo is a language model family from Ai2 for researchers and developers who want to run models on their own hardware or study how they're trained. It provides model weights, datasets, code and reports for examining and adapting the models.

Stable Diffusion XL is an image generation model for creators and developers who want to produce images on their own hardware or servers. You can start with a text prompt or use an existing image as input.

Needle is an AI model for developers building device assistants that turn requests into actions. It prioritizes automation over general chat, with smaller model variants for devices that have limited resources.

MuseTalk is a local AI lip-sync model for video creators and developers building virtual avatars. It edits a face to match supplied speech, so you can dub existing footage or animate a portrait.

Kokoro is a text-to-speech model and inference library for developers adding speech generation to personal projects or production applications. It runs on your own hardware and uses the compact Kokoro-82M model.

Uni MoE is a family of multimodal AI models for researchers working with language, visual material and audio together. Its Omni model can reason across combined inputs and answer questions about their content.

Jamba is AI21's language model family for teams building document-based assistants and business applications on their own servers. The Large model targets lengthy context, but its hardware requirements put it in server territory: quantized deployment uses eight GPUs with 80GB of memory each.

HiDream-I1 generates images from text prompts for creators and developers who want to work on their own hardware or add image generation to an application.

GTE is Alibaba's family of text embedding models for developers building semantic search and text matching into applications. The models turn text into vectors that let an application compare queries with documents by meaning.

Janus-Pro is a multimodal AI model for developers and researchers who want to interpret images and generate pictures on their own hardware. It combines both tasks in one model.

TripoSG generates 3D meshes from visual references for people creating 3D assets or testing shape ideas. It accepts finished images as well as rough scribbles paired with a text prompt.

DINOv3 is Meta AI's family of vision foundation models for researchers and developers building image analysis systems on their own hardware. Its pretrained models extract features for individual image regions that can support multiple vision tasks without fine-tuning the backbone.

Jina Embeddings is a family of models for developers building semantic search, retrieval-augmented generation (RAG) and AI agent applications. The models convert content into vectors that a search system can compare for relevance.

Depth Pro estimates distances from a single image for researchers and developers who need depth maps with an absolute scale. It can make predictions without camera metadata, so images don't need to come with known camera intrinsics.

MusicGen is Meta AI's model for generating music from written descriptions. It's part of AudioCraft, a PyTorch library for developers and audio researchers building music generation tools in their own computing environment.

GLM-4.5 is an open-source language model family for developers building AI agents and coding tools on their own servers. It supports reasoning and tool use, but its supplied inference configurations require server-class NVIDIA hardware.

VoxCPM is a local text-to-speech system for developers and audio creators who want to generate speech with voices they can design or clone. It supports multilingual narration and control over how a voice delivers the text.

Open Sora Plan is a video generation project for researchers and developers who want to work with model checkpoints on their own hardware. It aims to reproduce capabilities of OpenAI's Sora, with text-to-video and image-to-video models.