We use cookies. Cookies and browser storage keep you signed in and remember your preferences. With your permission we also measure how the site is used. You can any time. See our cookie policy.
amazon/titan-embed-image-v1Runs on your balance at the prices above. New accounts start with free usage.
Settings
Amazon Titan Multimodal Embeddings G1 is an embedding model from Amazon, listed on api14 under the model id amazon/titan-embed-image-v1. It sits in the embeddings category, which means its job is to turn inputs into vectors rather than to generate text or images. It is aimed at developers building search and recommendation systems that need to work across more than one type of content.
The model accepts text, images, or text and image together, and produces embeddings as its only output. Because all of those inputs are mapped into a shared vector space, a text query and an image can be compared directly, which is what makes cross-modal retrieval possible. The context window is 128 tokens, and the maximum output is 115. Images are supplied as base64.
Those capabilities point to a fairly specific set of uses. Multimodal search is the obvious one: a user types a description and the system retrieves matching images, or a user supplies an image and the system retrieves matching text. Recommendation is the other main case, where items that mix images and text are embedded and compared by similarity. Any workflow that needs a single vector index spanning both modalities, rather than two separate indexes, fits this model.
You call it through api14's OpenAI-compatible API using the model id amazon/titan-embed-image-v1, so existing OpenAI client libraries work without changes. The model supports no additional request parameters beyond the standard ones for the endpoint. Since the context window is 128 tokens, keep text inputs short and treat each request as a single embedding operation rather than a long document to be chunked by the model itself.
If your inputs are long passages of text, the 128 token window will be the binding constraint, and a text-only embedding model with a larger context is likely a better fit. The same applies if you need generated output rather than vectors, since this model only returns embeddings. For short text, images, or a combination of the two that needs to live in one shared vector space, it is the appropriate choice.
import OpenAI from "openai";
const client = new OpenAI({
baseURL: "https://api14.com/api/v1",
apiKey: process.env.API14_API_KEY,
});
const result = await client.embeddings.create({
"model": "amazon/titan-embed-image-v1",
"input": [
"The quick brown fox",
"Lorem ipsum"
]
});
console.log(result.data.map((d) => d.embedding.length));
This community has no topics yet. Open the first one about Amazon: Titan Multimodal Embeddings G1.