EmbeddingGemma 2: Search text, images, audio and video in one shared space
Added · Updated
Made by Google
View profile on Hugging Face (opens in a new tab)A 740-million-parameter embedding model from Google DeepMind that turns text, code, images, video and audio into vectors in one shared 768-dimensional space.
That lets an app search across different media, find similar items, group a collection, or retrieve evidence for a question without using a separate model for each format. The card says the text backbone can be loaded on its own at 270 million parameters, while the vision and audio encoders are optional.
Model details
Original model google/embeddinggemma-2
Input Text, code, image, video or audio Output 768-dimensional embedding
Ways to use it Run locally
Maps four media types into one shared space for search and retrieval.
What it does
An embedding is a list of numbers that captures what an input means, so two things with similar meaning land near each other when an app compares them. EmbeddingGemma 2 puts four kinds of input — text and code, images, video and audio — into the same 768-number space. A product catalogue can therefore match a written search with a product photo, a meeting archive can retrieve the video moment that answers a question, and a knowledge base can mix documents with diagrams. For text-only work you can omit the larger encoders; for storage-sensitive search the card documents shortening the vector to 512, 256 or 128 dimensions, with a quality tradeoff. Short task instructions tell the model whether it is answering a search query, representing a document, clustering or checking a claim.
You could use it to…
- Search a product catalog with a written query and match product photos
- Find the video moment that answers a question about a meeting archive
- Group a collection of similar documents and images
Can I use this?
You'll need Python with Transformers or Sentence-Transformers, and suitable local hardware · Setup needed
Worth knowing The model card says quality varies by language and task, and recommends task prefixes for text; it also warns that float16 can produce bad or unusable embeddings, so use bfloat16 where supported or float32. Google reports the benchmark results in the card, not BuildTube. Apache 2.0 plus Google’s Gemma Prohibited Use Policy.
Why it's here
Not trending this week
Last seen on the Hugging Face trending list on 27 October 2026. It stays on BuildTube so you can still find it.
- 3,194 people have liked it on Hugging Face.
- It was downloaded 507,451 times in the last 30 days.
Numbers checked on 8 November 2026; not refreshed since.
Behind it
View model pageBuildTube has not run or verified this project. Everything above is written from what the creator published.
Made this?
Quick questions
- What is EmbeddingGemma 2?
- An embedding is a list of numbers that captures what an input means, so two things with similar meaning land near each other when an app compares them.
- What can I use EmbeddingGemma 2 for?
- Search a product catalog with a written query and match product photos; Find the video moment that answers a question about a meeting archive; Group a collection of similar documents and images
- Can I run EmbeddingGemma 2 on my computer?
- It runs on your own hardware with a model runtime. Python with Transformers or Sentence-Transformers, and suitable local hardware.
- What do I need to use EmbeddingGemma 2?
- Install Python packages such as `sentence-transformers` and `transformers`, download the model, and choose the encoders your workload needs. The card is designed for consumer hardware, including laptops and mobile devices, but actual memory and speed depend on the modality combination and numeric precision.
- What license does EmbeddingGemma 2 use?
- The source lists the Apache License 2.0 license (Apache-2.0), an OSI-approved open-source license. Read the license file on the source page for the full terms.
More from this week
- Idol Turn one picture of an object into a 3D Gaussian splat
Idol turns one photo, render or generated image of an object into a 3D Gaussian splat file.
- P-H3-Image-Comfy Make and edit 8-megapixel images in ComfyUI
Turn short text requests into 8-megapixel images and edit pictures on your own graphics card.
- agent-desktop Let AI agents read and click desktop apps
For developers building computer-use agents: agent-desktop lets an AI agent read and operate desktop apps on Mac, Windows or Linux from a command line.
Explore by topic
Post this
Download share imageFor makers
Made this? Show it’s on BuildTube — add the badge to your README or embed the card on your site.
README badge (Markdown)
[](https://buildtube.app/p/hf-google-embeddinggemma-2/)Embed card (HTML)
<iframe src="https://buildtube.app/embed/hf-google-embeddinggemma-2/" width="560" height="240" style="border:0;max-width:100%" loading="lazy" title="EmbeddingGemma 2 on BuildTube"></iframe>