Microsoft just dropped TRELLIS.2, a 4 billion parameter image-to-3D model that generates fully textured 3D assets from a single image. It outputs radiance fields, Gaussian splats, and meshes. The geometry is sharp, the textures are accurate, and the whole thing runs at a scale that would have seemed impossible two years ago.
The research community is paying attention. Designers are paying attention. And they should be.
But here is the part nobody is talking about: generating a 3D asset is only step one.
What TRELLIS Actually Does
TRELLIS uses something called Structured LATent representation (SLAT), which lets it encode complex 3D structure into a compact form that a diffusion model can learn from at scale. The result is high-fidelity assets with complete textures, PBR materials, and the ability to handle complex geometry including sharp edges and detailed surfaces.
You give it an image. You get back a 3D asset in the format you need. Radiance field for real-time rendering, Gaussian splat for navigable scenes, mesh for traditional pipelines. The model handles the hard part.
It also supports local 3D editing, which means you are not locked into the initial generation. You can adjust parts of the asset after the fact, something previous models could not do cleanly.
The benchmarks back it up. TRELLIS significantly outperforms previous methods at similar scale, including models that were considered state-of-the-art six months ago.
The Gap Nobody Is Closing
Here is the honest problem.
TRELLIS is a research model. Deploying it requires a GPU, a Python environment, familiarity with the GitHub repo, and time to figure out how to connect its output to whatever you actually want to do with the asset.
For developers and researchers, that is fine. That is the intended audience.
For creative professionals, including product designers, advertising agencies, filmmakers, and content studios, that gap between "model exists" and "I can use this in my workflow today" is still enormous.
The same asset that TRELLIS generates beautifully still needs to be imported into a scene, lit, positioned, captured from specific camera angles, fed into a video generator, and assembled into final output. Each of those steps currently lives in a different tool, a different application, or a different command-line interface.
That is not a creative workflow. That is a technical project.
What No-Code Actually Means for 3D
The promise of no-code in creative tools is not about removing complexity. The complexity of 3D generation, scene composition, and video production is real and it has to live somewhere.
The promise is about where that complexity lives. Does it live in terminal windows and configuration files, or does it live in a canvas where a designer can see what they are doing?
Node-based creative workflows represent the middle ground that the industry has been missing. You get the power of the underlying models without needing to be the person who deployed them. You connect outputs to inputs visually. You see the result. You adjust. You move on.
When TRELLIS generates a 3D Gaussian splat, that asset needs a home. A place where it can be imported into a scene, where a camera can move through it, where captures can be taken at any angle and fed directly into the next generation step. A place where the whole pipeline from image to final video lives in one place and runs without switching between five different applications.
Tools like Raelume (raelume.ai) are building exactly that. A node-based canvas where 3D Worlds blocks let you import a Gaussian splat, move a camera through the scene, snap 4K captures from any angle, and feed those directly into video generation, all without leaving the canvas. TRELLIS handles the hard reconstruction step. The workflow handles everything that comes after.
That is the gap that no-code 3D workflow tools are filling right now, and TRELLIS arriving at production scale makes that gap more relevant, not less.
The Timing Is Not a Coincidence
Three major 3D generation tools launched or updated in March 2026 alone. TRELLIS.2 from Microsoft. Tripo P1.0 from Tripo AI, announced at GDC with enterprise-grade performance. Wonder 3D from Autodesk, bringing 3D generation directly into Flow Studio.
The pattern is consistent. Every major player in the space is pushing toward production-grade 3D generation. The models are getting better faster than anyone expected.
What that means practically is that the bottleneck is shifting. The hard part used to be generating a usable 3D asset at all. That problem is largely solved at the research level and increasingly solved at the production level.
The new bottleneck is workflow. How do you take that asset and turn it into something a creative team can actually deliver?
What This Looks Like in Practice
Consider a product photography workflow. A brand wants to show a new product in multiple environments, from multiple angles, without a physical shoot.
The old approach required a 3D artist to model the product, a rendering pipeline to place it in environments, and a post-production team to make it look real. Time: weeks. Cost: significant.
With models like TRELLIS handling the 3D reconstruction from reference images, the asset generation step collapses from days to minutes. But that asset still needs to go somewhere. It needs a scene. It needs camera positions. It needs to generate final output in formats the brand can actually use.
A node-based workflow that connects image generation, 3D reconstruction, scene composition, camera capture, and video generation in a single canvas turns that entire pipeline into something a single creative can run. Raelume is one of the few tools doing this end to end today, with Worlds blocks that handle the full journey from image to navigable 3D scene to final video output. The complexity is real but it is managed. The output is professional. The time is measured in hours, not weeks.
TRELLIS makes the 3D generation step better. No-code creative workflows make the entire pipeline accessible.
Those are different problems, and they are both being solved right now.
The Takeaway
Microsoft proving that 3D generation works at scale is genuinely significant. It validates what a lot of people in the creative industry have been building toward: a world where generating a high-quality 3D asset from a reference image is a routine step in a production pipeline rather than a research project.
But the creative professionals who will actually benefit from that breakthrough are not the ones running models from a terminal. They are the ones working in tools that put the model's output into their hands without requiring them to become ML engineers first.
The no-code moment for 3D creative workflows is arriving at exactly the right time.
Alex Mercer covers AI creative tools and workflows independently. No affiliation with the tools mentioned.
Top comments (0)