AI models · new in Prism 1.1.7

Download GGUF and safetensors models from Hugging Face — fast and verified

Search for a model, type its name or paste its link. Prism picks the files you need — one quantization of a GGUF model, or the weights in one format with the configs and tokenizer beside them — downloads each over many connections, and checks every weights file against the SHA-256 Hugging Face lists.

Download Prism How it works

Windows · macOS · Linux — 64-bit

Prism's Hugging Face queue: a 15.7 GB GGUF file downloading over 16 connections at 24 MB/s, with the model's configs, tokenizer and weights already completed. The Hugging Face queue in the light theme.

How it works

From a model name to verified files.

No Python, no command line: the model window is part of the download manager.

Step by step

  1. Open the model window

    Click the model button in the toolbar, or paste a model’s name like Qwen/Qwen3-8B — or its link — into Add URL. A link to a model’s page opens the model window; a link to a single file’s page downloads that file.

  2. Pick a version

    For a GGUF model, each quantization is listed with its size, and Q4_K_M is picked unless you choose another; a vision model’s mmproj file comes along. For other models Prism takes the weights in one format (safetensors when there are any) with the configs and tokenizer.

  3. Download

    The files go into the Hugging Face queue and download one after another, each over many connections. Files you already have are skipped.

  4. Verified

    Every weights file is checked against the SHA-256 that Hugging Face lists for it.

Searching Hugging Face for qwen3.8 in Prism: matching models with GGUF tags and download counts. Searching Hugging Face in the light theme.
Search right in Prism.
The model window for a GGUF model: versions Q4_K_M, Q6_K and Q8_0 with their sizes, Q4_K_M and the mmproj file selected, 412 GB free. The model window in the light theme.
Pick a quantization; Prism picks the files.

Details

Made for big model files.

Gated and private models

Add an access token in Settings › AI models, or let Prism use the one hf auth login saved. The token goes only to Hugging Face, never to the servers the files come from.

Mirrors

Download from a mirror such as hf-mirror.com, or whatever HF_ENDPOINT points to.

Straight into LM Studio

Files go to Downloads/Models/<owner>/<model>, or one click saves them into LM Studio’s models folder.

Slow download? More connections

Each file is split across up to 32 connections, which helps most when a single connection is slow or limited.

Resume 20 GB files

Pause tonight, continue tomorrow, survive a crash: Prism resumes from the last byte it safely wrote.

Overnight, on schedule

The Hugging Face queue is a queue like any other: give it a schedule and a speed limit.

Download

Free. For every desktop.

The model downloader is part of Prism 1.1.7 and later, on Windows, macOS and Linux.

Your system

macOS

macOS 11 Big Sur or later

First run: the app isn’t notarized by Apple yet, so macOS blocks it once. Run xattr -dr com.apple.quarantine /Applications/Prism.app in Terminal, or use System Settings → Privacy & Security → Open Anyway. Not sure which Mac you have? Apple menu → About This Mac → Chip.

Browser extension included: the DMG has a Browser Extension folder next to Prism — drag it to Documents, then load it in Chrome (Prism guides you on first start).

FAQ

Hugging Face downloads, answered.

Prism isn’t affiliated with Hugging Face.

How do I download a model from Hugging Face with Prism?

Click the model button in the toolbar (or paste a model’s name like Qwen/Qwen3-8B, or its link, into Add URL). Pick a version if the model has several, and click Download: the files go into the Hugging Face queue.

Which GGUF file does Prism pick?

One quantization: Q4_K_M unless you choose another version in the model window, plus the model’s mmproj file when it has one. You can also tick files by hand, or select all.

Why is my Hugging Face download slow?

Often because one connection is slow — the route to the server, a proxy, or a limit per connection. Prism downloads each file over several connections at once, which helps in exactly those cases, and a big file that gets interrupted continues instead of starting over. If you already download at your line’s full speed, nothing can go faster.

Can I download all the files of a model?

Yes. In the model window, Select all ticks every file in the repository, and Recommended goes back to Prism’s pick. You can also tick files one by one.

Can it download gated or private models?

Yes, with an access token: add one in Settings › AI models, or Prism uses the one hf auth login saved. The token is sent only to Hugging Face, never to the servers the files are redirected to, and it isn’t stored with the downloads. For gated models, accept the model’s terms on huggingface.co first.

Does it work with hf-mirror.com?

Yes. Choose a mirror such as hf-mirror.com in Settings › AI models, or set HF_ENDPOINT.

Does Prism check the downloaded files?

Yes. Every weights file is checked against the SHA-256 Hugging Face lists for it, and files you already have are skipped.

Can LM Studio use the models Prism downloads?

Yes. When LM Studio is installed, the model window offers Save into LM Studio’s models folder, and the files land where LM Studio looks for them.

More about Prism

The Prism home page has the full feature tour, benchmarks and the install guide.

Prism homeFeatures, benchmarks, film and guide. Prism for WindowsInstaller, MSI or portable for Windows 10 and 11. Prism for MacApple Silicon and Intel, macOS 11 or later. Prism for Linuxdeb, rpm or AppImage for any distribution. Browser extensionChrome, Edge, Brave, Opera, Vivaldi and Firefox. An alternative to IDMWhat carries over from Internet Download Manager.

One beam. Many rays.

Free, open source and ready for your next big download.

Download Prism Star on GitHub