Amazon Titan Multimodal Embeddings G1 model - Token Usage (bedrock)

Thank you for your questions about the Amazon Titan Multimodal Embeddings G1 model. I'll address each of your points:

a) Currently, there is no direct way to return the image token usage for the Amazon Titan Multimodal Embeddings G1 model (amazon.titan-embed-image-v1) through the API response. The API only provides the inputTextTokenCount, which applies to the text input. The image processing is handled differently and is not measured in tokens in the same way as text.

b) You're correct to question the pricing structure for the multimodal embeddings model. The pricing for the Amazon Titan Multimodal Embeddings G1 model is indeed different from the Amazon Titan Image Generator model.

For the Multimodal Embeddings model, the pricing is set at a flat rate per image, regardless of the image size. This is likely because the model processes the image to create a fixed-size embedding, rather than generating a new image. The flat rate simplifies the pricing structure for this particular use case.

In contrast, the Amazon Titan Image Generator model's pricing does vary based on the output image size and quality, as you correctly noted. This is because generating images of different sizes and qualities requires different amounts of computational resources.

It's important to note that while the pricing for the Multimodal Embeddings model appears to be independent of image size, there are still limitations on the maximum input image size (25 MB) that the model can process.

The pricing structure for each model is designed to reflect its specific use case and the computational resources required. If you have concerns about the accuracy of the pricing information, it would be best to contact AWS support directly for the most up-to-date and accurate pricing details.

Sources
Amazon Titan Multimodal Embeddings G1 model - Amazon Bedrock
Amazon Titan Multimodal Embeddings G1 - Amazon Bedrock
Amazon Titan Image Generator G1 models - Amazon Bedrock