Upcoming Inworld product launches and roadmap
During our recent product roadmap webinar, we presented new Inworld features that enhance the believability of AI agents, simplify creative workflows, and engage with broader audiences.
If you missed the webinar, you can view the recording or continue reading for a summary of the features we covered.
Goals, Activations, Actions and Mutations
Last year, we introduced the Mutations function as part of our Goals and Actions system to enable dynamic changes to core character attributes during runtime. In our latest updates, we expanded the Mutations function to offer greater control over all character fields during runtime, allowing temporary mutations in production.
Developers now have new creative possibilities for temporary character transformations, thanks to over 30 mutable fields available for modifying a character's brain during gameplay. Take a look at our Mutations function in our Inworld Unreal Playground.
Real-Time Data Injection
Our new Real-Time Data Injection function allows you to create more dynamic and interactive player experiences with content that adapts to real-world information. Our user-friendly system lets you embed live data directly in the character fields and design functions in your client-side logic that directly impact AI characters' narratives. You can test this functionality for yourself in our Inworld Unreal Playground.
Multilingual Support
To help our customers access new player markets, we have made substantial investments in Multilingual Support. Unlike traditional translation methods, this feature takes into account the unique cultural contexts of different languages, such as colloquial phrases, proverbs, and even safety control considerations that can significantly differ across languages. We achieved this by incorporating support directly into our entire stack, enabling a seamless end-to-end experience across all languages. This integration extends to fundamental components like knowledge retrieval and character dialogue generation.
We are currently in alpha with Chinese, Korean and Japanese Multilingual Support, with major European languages coming soon.
Feedback Loops
Our Feedback Loops feature was designed as a useful tool during the character testing and iteration phase as a means to fine-tune characters. It gives you the ability to provide useful dialogue quality feedback through the simulator, allowing them to address various aspects such as preferred conversation styles, fact corrections, tone adjustments, and safety considerations.
By incorporating this customer feedback, we are able to continuously retrain our model, leading to significant success in fine-tuning specific dialogue styles. This feature will be available to select partners.
Multi-Character
Our new Multi-Character functionality significantly expands the narrative possibilities in gameplay. By giving developers the ability to create multiple characters in a scene, we allow characters to have their own conversations independent of the player. This opens up opportunities for more intricate relationships, a range of emotions, and engaging dialogue.
In our watch mode, characters engage in free-flowing conversations, creating plot points and developing their motivations with each other without any player intervention. This creates a natural self-generating dialogue between the characters.
In our chat mode, players have the power to interject and guide the conversation. The characters respond to the player's presence and can be specifically called upon for interaction. Players can introduce themselves or direct the conversation towards a particular character, and that character will respond accordingly. Stay tuned for more information about Mutli-Character coming soon.
Hallucination Control/Knowledge Filters
Our Knowledge Filters feature addresses one of the most common game developer concerns - hallucination. This feature adds parameters that not only address the traditional issue of incorrect facts but also tackles the challenge of maintaining immersion when dealing with potentially true facts that can disrupt the overall experience.
Our Knowledge Filters feature offers two modes for enhanced control. In mild control mode, characters refrain from projecting excessive knowledge, avoiding providing Wikipedia-like answers to scientific questions beyond their realistic scope.
In strict control mode, characters avoid fabricating details, making it ideal for games where precise interpretation of facts is crucial, such as murder mysteries. In this mode, characters refrain from introducing narrative-altering details in response to knowledge-related or unrelated questions, ensuring they do not create any new facts.
Voice
Inworld’s new Voice capabilities introduce more lifelike and more natural-sounding voices into our Studio that add emotional depth to character interactions. For no additional cost, Inworld customers now have access to 48 new high-quality voices. Improvements from our previous Studio voices include lower latency, improved rhythm, intonation, pitch variation, and pausing that ultimately contribute to more realistic interactions. These new voices are now available to our customers in the Inworld Studio.
Our new voice cloning service allows customers to create voice clones for specific characters or individuals. By providing just five minutes of audio data from hired voice actors, developers and content creators can bring their characters to life with authentic and lifelike voices. This service is available for select customers only.
Reasoning Step
The Reasoning Step enhances characters in your game by enabling reasoning skills alongside verbal expression. This feature serves as a flexible prototyping ground, eliminating the need for manual logic encoding. This feature allows for various character complexities and enables the ingestion of game state for additional character reasoning. It also facilitates state reasoning to maintain continuity across different time periods. The Reasoning Step is available for select customers as part of our Inworld Core.