Release notes for Inspire: August 12th, 2024

🚀 Multi-Modal Support - Use Images as Input

Generate Text from Images:

Clients can now leverage images as input for text generation. This innovative feature allows you to create more accurate and detailed product descriptions by combining PIM information with visual content. Whether you want to describe a product by analyzing its image or use prompts based on visual data, this feature enhances the richness and precision of your product descriptions.

Extract Information from Images:

Another powerful use case is the ability to read text or extract detailed information directly from images and save it into designated fields. This streamlines the process of capturing and utilizing visual data within your content workflows.

Seamless Integration with New Models:

This feature is fully compatible with the new models GPT-4o-mini and GPT-4o, allowing you to take full advantage of the latest advancements in AI. These models enhance the accuracy and relevance of the generated content, especially when using images as input, ensuring high-quality product descriptions and content creation.

This new feature opens up a world of possibilities for enhancing your product content, making it more detailed, accurate, and visually informed.

Note: that using images as input will consume more input tokens than just using text as input. 

Flexible Image Input Configuration:

Image input supports a wide range of formats, including jpg, jpeg, png, gif, and webp. You have multiple configuration possibilities, from fetching images directly from the current entity to more advanced setups like retrieving images from sub-entities, such as items (variants), or from specific fields. This flexibility allows you to tailor the image input to suit your specific needs and workflows.

Advance example

🚀 Enhanced Token Efficiency with GPT-4o-mini

Maximize Your Token Usage:

With the introduction of the GPT-4o-mini model, all clients can now use a lot more tokens each month, maximizing the value of their subscription. By default, each inriver Inspire client is allocated 1 million tokens per month. However, by leveraging GPT-4o-mini, you can dramatically extend the reach of these tokens.

Reduced Token consumption:

Input Token consumption: Using GPT-4o-mini reduces the input token cost to just 1/10th of the standard model (GPT-3.5 turbo). This means you can process ten times more input data with the same number of tokens.

Output Token consumption: The output token cost is reduced to 1/3rd of the standard model (GPT-3.5 turbo), allowing you to generate three times more content with the same token allocation.

Example of Token Efficiency:

Standard Model (GPT-3.5 turbo): With 1 million tokens, you might generate 200,000 input tokens and 800,000 output tokens.

With GPT-4o-mini: The same 1 million tokens could now accommodate 2 million input tokens (10x increase) and 2.4 million output tokens (3x increase), effectively multiplying the amount of content you can generate by several times.

Gradual Transition to GPT-4o-mini:

We will gradually transition all clients from the GPT-3.5 turbo model to the GPT-4o-mini version. This move is designed to provide each client with significantly more tokens and enhanced capabilities, allowing for even more detailed and extensive content creation. Additionally, this transition paves the way for the integration of improved LLM models in the future, ensuring that you benefit from the latest advancements in AI technology.

Why GPT-4o-mini?

GPT-4o-mini is designed for efficiency without compromising on quality. It’s ideal for clients who need to manage large volumes of content while keeping costs low and output high.

This new feature not only enhances your productivity but also ensures you get the most out of your inriver Inspire subscription.

🚀 Sub Configuration

Fine-Grained Prompt and Input Configuration:

Inspire admin users can now enjoy more precise control over prompt and input configurations, enabling an optimal experience tailored to specific needs.

Enhanced Text Generation for Product Categories:

This feature allows you to support highly specific text generation for products across different categories. Whether it’s a luxury brand or a budget-friendly line, you can now craft content that perfectly aligns with the category’s identity.

Improved Multi-Brand and Multi-Segment Support:

Sub Configuration allows for the creation of differentiated prompts for various brands, segments, or a combination of both. This ensures that each brand or product category receives the attention it deserves, with prompts tailored to its unique voice and tone.

Brand and Category-Specific Writing Instructions:

You can now integrate brand or category-specific instructions directly into the prompts and use different input field and prompts per sub configuration. This means you can maintain a consistent tone and brand voice, ensuring that your generated text reflects the nuances of your products, whether they belong to different brands or categories.

Copy Default Sub Configuration:

To speed up the creation of more specific sub-configurations, you can easily copy the default sub-configuration. This allows you to quickly customize and fine-tune configurations without starting from scratch, streamlining the process of setting up detailed and tailored prompts.

This new feature empowers you to create more relevant, on-brand content, enhancing the overall quality and effectiveness of your marketing texts.

🚀 Re-Generate Text in Inspire Batch Review Process

Enhanced Review Flexibility:

Users can now re-generate text for specific fields during the review process. If the initial batch-generated text doesn’t meet your expectations, you can easily request our AI to generate a new version tailored to your needs.

Streamlined Content Editing:

This new text can then be reviewed, accepted, and used in place of the original batch-generated version. This feature empowers you to fine-tune and perfect your product descriptions and other content before it’s finalized and saved to the inriver PIM.

Greater Precision and Control:

This enhancement to the review process ensures that the final text aligns with your brand’s tone, voice, and accuracy standards, offering a more precise and customized approach to content creation.

This new feature enhances the flexibility and precision of your content workflow, ensuring that you always have the option to refine and perfect your text before it goes live.

🚀GPT-4o support

Inspire now supports the advanced GPT-4o model, enabling more sophisticated text generation. However, be mindful that GPT-4o consumes 6x more tokens than GPT-3.5 turbo, so for most use cases, we recommend using GPT-4o-mini, which balances advanced capabilities with efficient token usage. This upgrade allows for higher quality and more precise content creation, giving you access to cutting-edge AI technology in your workflows.

🚀Client specific system prompt and prompts

Prompts can now easier be accessed by all Inspire Admin users direct from the configuration modal.

🚀Smaller improvements

Regexp support - If a field has Regexp rules in Enrich, then these rules will also be shown and used in inspire and inspire Batch.

Improved configuration - more descriptions related to if field are not added. 

Improved backup - process on backup is now visualized for the Admin users of the configuration.