Hi everyone, I am excited to share our Sentis roadmap with you . Our goals with this are:
1. Transparency: Share the “What?” and “Why?” we are building. The “How?” will come with actual releases because it often changes as features mature. As we gaze toward a production release with Unity 6 around October 2024, we will be as transparent as possible with what is coming.
2. Feedback: We’d like your feedback/votes on our posted ideas. If you can provide specific technical requirements aligned to use cases, that context helps us understand your needs.
3. New ideas: Solicit new and fresh ideas that we have yet to think of. This is my favorite part .
A little birdy on my shoulder just tweeted that 3 of the features currently “In Progress” on the roadmap are coming in a new release around the end of next week … so stay tuned for that!
Use case: replace hand baked AI with something that mimics the desired behavior.
Interaction: player 1 is the protagonist, player 2 controls one of antagonists or 2 neural nets play each side.
Result: the antagonists can now play against a player or survive.
Format: attributes to tag inputs and maybe even properties that score the agent.
I don’t think this has any effect on Sentis. Sentis is just a tool for importing, optimizing, and running pre-trained neural networks; it does not come with any neural networks installed.
Also, the vast majority of use cases that Unity developers are using are for very specific game features and real-world interactions, for which the source models are widely available and trained on open/free data sets, or the models are built directly by the game developers with private/owned data sets. These use cases and models generally do not overlap with highly contentious, legally grey models like SD. However, if you are trying to run SD on Sentis specifically, it will affect you.
yeah sorry mixed up with that other unity AI system - with all these rebrandings : )
(i think their blogpost said it used SD in some parts of training “minimally”…)
Is there any timeline for just dropping in well-known models and running them? (without onnx converting or installing python and dealing with mountains of trouble)
Really just want to drop in some quantized LLM models from hugging face or some flux/stable diffusion models and just get going. It’s still just simpler to run a local web server and offload requests and get back text/image. Would be nice to be all integrated and just run locally with all the burst/threading/performance control that sentis can provide by being integrated with unity.
. Our goals with this are:
.