AI Assisted Search in Cinema 4D 2025.3

Doug Y
Doug Y
  • Updated

Understanding Our New AI-Powered Search Feature

We recently introduced an exciting new feature to the Asset Browser that uses locally run artificial intelligence (AI) to enhance search functionality for assets. This article explains how the system works, highlights the measures we've taken to ensure your privacy and security, and provides options for managing the feature to suit your preferences.

How It Works

Our AI-powered search uses a sophisticated model trained on a large dataset called WebLI. This dataset consists of a vast collection of image-text pairs sourced from publicly available web content. It is designed to help the AI model understand connections between images and text, enabling it to perform tasks based on meaning rather than literal keywords. You can learn more about WebLI here.

What Are Embeddings?

An embedding is a compact, abstract representation of an image or piece of text. It acts like a mathematical fingerprint, allowing the model to analyze and compare how similar different embeddings are. This capability powers semantic search, which retrieves results based on meaning rather than exact matches.

For example:

  • Searching for "chair" might also display images of barstools because, conceptually, both are types of seating.
  • Searching for "woman carrying bag" versus "women carrying bags" will reflect the subtle difference in your query, showing how nuanced the system can be.

Beyond Semantic Search

The AI model has also been trained to perform basic OCR (Optical Character Recognition), allowing it to recognize and search for text within images. For example:

  • Searching for "riesling" might return images of wine labels.
  • Searching for "woodnex" could surface a photo of a wooden product with that word printed on it.

Additionally, the system supports image similarity search, where you can provide an image and find other images that are visually or conceptually similar. For example, this feature can help you locate alternate textures or styles for the same object.

Why Results Might Not Always Be Perfect

When comparing embeddings, the system measures their "distance" from one another. Closer distances mean higher similarity. However, because the model works at a conceptual level, you might occasionally see results that feel slightly off.

For example:

  • Searching for "bird" might eventually return an image of a squirrel. Both are animals, and the system may include results that are conceptually similar when more precise matches are unavailable.

Privacy and Security

We understand that privacy and security are critical to our users, which is why this feature has been designed with these principles in mind.

  • Local Processing: All AI processing happens locally on your machine. This means your images, text, and embeddings never leave your system. When you perform a search, the application computes embeddings locally and sends the results back to you without transmitting any data online.
  • Precomputed Embeddings for Asset Library: For assets provided by us, embeddings are precomputed to ensure faster searches. If you add your own assets to the Asset Browser, the system computes embeddings for those items in the background, enabling real-time search functionality.
  • User Control: While we believe this feature is safe and beneficial, we understand that some users may prefer not to use it. You can disable AI processing entirely in the Cinema 4D’s preferences (Asset Browser section). In any case, no data or embeddings are shared externally, and the AI model does not use your assets or inputs for further training.

Hardware Utilization

The AI model takes advantage of your system’s hardware to deliver fast and efficient performance:

  • Windows: The system uses NVIDIA GPUs, with experimental support for AMD and Intel GPUs.
  • macOS: On devices with M-series chips, it utilizes the Neural Engine built into Apple’s SoC architecture.

These optimizations ensure the feature runs smoothly while keeping all processing local to your machine.

 

We hope this article clarifies how the AI-powered search works and provides transparency about the measures we've implemented to protect your data. If you have further questions or feedback, feel free to reach out to us.

Was this article helpful?

/

Comments

0 comments

Article is closed for comments.