Multimodal models allow interactions that go beyond traditional text-based prompts. These models can understand and process information from a variety of sources, including text, file documents, images, audio and video. This means they can analyze, process and generate content based on a richer and more complete set of inputs.
GeneXus Enterprise AI implements multimodal models, allowing you to interact with models such as Gemini and OpenAI in a completely new way. By combining text with files, a world of possibilities opens up for analyzing, processing and generating content.
These models can directly receive files alongside your prompts, eliminating the need to manually extract and inject file content.
This direct file integration offers several advantages.
- Simplifies file handling. You no longer need to extract and inject file content into the prompt.
- Files are sent directly to the model, streamlining the process and enabling direct model interaction.
- Empowers you to interact directly with files within your prompts, making it easier to analyze, process, and generate content based on their contents. This eliminates the need for manual extraction and injection, simplifying your workflow.
- Leverages the information contained in files to provide more comprehensive and insightful responses, enhancing their capabilities.
You can reference files in your prompts using a specific format: {file:filename}. For example, to reference a file named "report.pdf" you would write: {file:report}.
Note: This reference format only works for files that have already been uploaded to your project. If you're uploading a file for the first time, you'll need to use the standard file upload mechanism within the platform.
- File Storage: All files uploaded to your project are stored securely and can be accessed by your prompts.
- File Access: When you reference a file in your prompt, the system automatically retrieves it and sends it to the multimodal model.
- File Types: The system supports a wide range of file types, including text documents, images, audio files, and more.
Suppose you have a document called "product_specs.txt" loaded in your project. You can use it in your prompt as follows:
"Please summarize the key features of the product described in {file:product_specs}".
The multimodal model will then access the "product_specs.txt" file and use its contents to generate a summary of the product features.
Supported models and file formats