How an Agent Built a 3D Paris Gallery by Chaining Two Hugging Face Spaces
Original reporting by Hugging Face
An AI agent, given a simple prompt, recently constructed a cinematic 3D gallery showcasing Paris monuments as Gaussian splats. What makes this feat remarkable is that the agent generated every single asset—from source images to 3D splats—without human intervention, by directly calling two distinct Hugging Face Spaces. It never touched an image generator or a 3D reconstruction tool manually; instead, it orchestrated existing, pre-trained components to deliver the final interactive experience, deployed live as a static Space.
The building block approach
This demonstration illustrates a profound shift in software development, aligning with what's called the "building-block economy." Instead of monolithic applications, the most effective path involves AI agents assembling small, well-documented components. While this concept has been discussed for code libraries, it is now rapidly transforming multimedia AI. Hugging Face Spaces, which host thousands of state-of-the-art models, have quietly become these standardized building blocks. Each Gradio Space now exposes an `agents.md` file, providing agents with all necessary API information in a plain-text format. This enables agents to seamlessly chain outputs from one Space as inputs to another, dramatically lowering integration barriers that once made "prompt-to-3D-splat" a complex project. This new paradigm heralds a future where sophisticated multimedia creation is a fluid conversation between human and AI, assembling powerful primitives with unprecedented ease.
The demonstration of an AI agent constructing a 3D gallery from distinct Hugging Face Spaces underscores a pivotal shift in multimedia development. By leveraging the standardized `agents.md` interface, specialized AI models — previously isolated by integration complexities — are transformed into universally callable building blocks. This radical simplification moves the bottleneck from intricate coding and setup to defining high-level creative or functional objectives, making sophisticated AI workflows accessible to a broader range of developers and non-technical users alike. The vision of a "building-block economy" for software, where agents excel at assembling proven components, is now manifesting concretely in the realm of generative AI and 3D reconstruction.
A New Paradigm Emerges
The implications of this composability extend far beyond novel multimedia galleries. This paradigm signals a profound acceleration in innovation, enabling rapid prototyping and iterative development where complex, multi-stage AI pipelines can be spun up in moments. It positions platforms like Hugging Face as vital ecosystems, curating a vast library of interoperable AI primitives that agents can dynamically discover and combine. In the near future, we can anticipate a proliferation of bespoke applications, autonomously generated content, and entire creative agencies orchestrated by AI, all built by intelligently chaining these increasingly powerful and accessible blocks. The era of the AI orchestrator, adept at assembling the best available tools to achieve human-defined goals, has truly begun.