We share thoughts on product development and future AI directions at InferenceCloud

Get the complete picture of what’s coming next from our Product Development Roadmap

(Actually useful) Image generation is coming

March 2025

Feature under development

We’ve been working on image generation for some time, but we’ve struggled to create a reliable solution that accurately depicts the requested content whilst maintaining brand compliance. It’s been hard to make images useful!

But new techniques in AI are changing that. A move from diffusion based image generation to auto-regressive models allows us to build images based on more specific data, allowing us to iterate on specific input images and brand examples.

Here is the concept that we are working on right now which brings the core approach of InferenceCloud to images (and video at some point in the future):

Defining your needs

First, Insight Miner (which now needs a new name) allows you to upload images as well as documents that define the type of content you need.

So, for example, you can upload images or renders of your product as well as images that depict the environment you want it depicted in, say, a shiny black box, and a busy office environment.

When we define organizational style guidelines (when setting up your InferenceCloud account) we can also upload a selection of brand images to define your corporate identify.

Using data to create high impact images

Strategy Explorer will also identify the nature of images associated with specific topics, specifically those which are strongly associated with intent, engagement and action.

Inference Writer (also needs a new name) will then combine the following data points to create a selection of images:

• Your brand guidelines
• Images of products, settings or other concepts from Insight Miner
• The key content details and intent from the text generated
• Guidelines on impactful images for the given topic from Strategy Explorer
• Any other style or specific requests that you have

The goal of this approach is to avoid the constant manual iteration that is often required for image generation with the ability to create useful and on-point images. Newer techniques allow us to iterate over images as they are generated so that we can evolve the perfect, useful image.

Place the product in different locations depending on post context, for example an emphasis on “sport and adventure”

… or an emphasis on life and a more cosmopolitan feel

Further adapt the perspective based on post or topic intent

For example, when content is in the context of family benefits

Goals for 2025

December 2024

We have a big year planned for communications AI in 2025. Here’s what’s cooking.

Our goal at InferenceCloud is to solve the biggest challenges in communications. To continue making this a reality we’ve set some fundamental objectives:

• Make our AI even smarter and more aware of the very specific needs of each customer
• Make AI significantly easier to use, whilst still maintaining human-in-the loop as a key part of the process
• Create output that is better and better
• Provide new insights with a focus on actionability and measurable results

Here are the things we are working on to achieve this in 2025:

Context agents

One of the big challenges with AI today is the lack of embodiment. AI doesn’t have access to all the small facts, details and trivia that inform a complicated process such as communication. - what was that issue mentioned on the news this morning? what did somebody say in a meeting yesterday?