Titan Multimodal Embeddings G1 (or any multimodal embeddings model for text/OCR/image/tables) available as an embeddings model when creating a new knowledge base in bedrock?

You need to enable the model in "Model Access". Did you do it?

I apologize, but there seems to be a misunderstanding in your question. The Amazon Titan Multimodal Embeddings G1 model is not currently available as an option for creating knowledge bases in Amazon Bedrock.

The Titan Multimodal Embeddings G1 model is designed for tasks like searching images by text, searching images for similarity, or combining text and image inputs. However, it is not specifically designed for or offered as an embeddings model for knowledge base creation in Bedrock.

For creating a knowledge base in Bedrock, especially one that needs to handle various types of content like PDFs with text, images, tables, and scanned text, you would typically use a text embeddings model. The available options for knowledge base creation are more limited and focused on text-based embeddings.

If you're looking to create a knowledge base that can handle mixed content types, you might need to consider a different approach:

1. Use a text embeddings model that's supported for knowledge base creation in Bedrock. This will work well for the text content in your PDFs.

2. For images and tables, you may need to use separate processes to extract text or relevant information from these elements before adding them to your knowledge base.

3. For scanned text (OCR content), you would first need to use an OCR tool to convert the scanned images to text, and then add that text to your knowledge base.

Unfortunately, there isn't a single, out-of-the-box solution in Bedrock for handling all these content types in a knowledge base simultaneously. You may need to pre-process your PDFs to extract and convert all content to text before creating your knowledge base.

If you're specifically interested in multimodal capabilities, you might want to explore other AWS services or third-party solutions that specialize in handling mixed media content for search and retrieval purposes.
Sources
Amazon Titan Multimodal Embeddings G1 model - Amazon Bedrock
Create an Amazon Bedrock knowledge base - Amazon Bedrock
Model support by feature - Amazon Bedrock