SeamFlow Improves UV Unwrapping via Flow Matching on Edge Probabilities
Meshy released SeamFlow, a new UV unwrapping algorithm that applies flow matching to edge probabilities to place seams without autoregressive errors. 3D artists and game developers benefit from cleaner texture maps and more predictable UV layouts when exporting models.
Meshy, the AI 3D generation platform, released SeamFlow, a new UV unwrapping algorithm designed to improve how textures map onto generated 3D assets. The update reformulates UV seam placement as flow matching on edge probabilities, eliminating the projection errors and ordering biases common in traditional autoregressive cutting methods. This change ensures that exported 3D models have cleaner, more logical UV layouts for texturing.
What's new
The SeamFlow algorithm addresses a major bottleneck in automated 3D pipeline workflows: messy UV coordinates. Traditional AI-based UV unwrapping tools often rely on autoregressive models that decide where to cut seams sequentially. This sequential approach introduces ordering bias, where early decisions negatively impact later cuts, resulting in fragmented UV islands and distorted textures.
To solve this, Meshy developed SeamFlow to predict edge probabilities across the entire 3D mesh simultaneously using a flow matching framework. By evaluating the probability of every edge being a seam at the same time, the model avoids the compounding errors of step-by-step cutting. The result is a significant reduction in projection distortion and more cohesive UV islands that resemble manual unwrapping.
How it fits your workflow
For 3D artists, game developers, and VFX technical directors, UV unwrapping is often the most tedious part of the asset creation pipeline. While automated tools in Blender or Adobe Substance Painter offer quick unwrapping, they frequently produce highly fragmented maps that are difficult to edit manually. Meshy's SeamFlow provides a cleaner starting point directly out of the AI generation phase, making it easier to import assets into external digital content creation software.
In comparison to alternative AI 3D generators like Tripo3D or Luma Genie, which often output messy, dense UV maps requiring manual retopology and re-unwrapping, Meshy's integration of SeamFlow aims to deliver production-ready assets directly. This makes the platform highly competitive for rapid prototyping and background asset generation, where manual UV editing is too time-consuming.
What it costs / how to try it
SeamFlow is integrated directly into the Meshy 3D generation pipeline as of February 2025. Users can test the improved UV unwrapping by generating new 3D models or texturing existing meshes through the Meshy web application, which offers both free trial credits and premium subscription tiers.
Read the original announcement on Meshy ↗