503 Service Unavailable hitting multiple major Image-to-3D Spaces (TripoSR, InstantMesh, LGM) via Gradio Client

Thanks for the insight on the runtime errors. You’re right that the Spaces are currently in a crash state, but I suspect this is a broader infrastructure issue rather than individual code errors by the authors for a few reasons:

1. Statistical Anomaly: It is highly unlikely that four separate research teams (Stability AI, Tencent, etc.) all broke their unrelated codebases at the exact same hour yesterday, & today.

2. External Failures: I’ve confirmed that Microsoft Copilot Labs’ 3D generation is also currently failing. Since Copilot’s 3D features and TripoSR share the same architectural DNA (Tripo AI), this points to an upstream API or model weight server failure.

3. The ‘3D 2.0’ Migration: Tripo AI officially launched their ‘AI 3D 2.0’ (Smart Mesh P1.0) on April 1st. It’s very possible that a major deprecation of the v1 endpoints or a change in how model weights are served has orphaned these older Spaces.

4. Environment Drift: The logs across these Spaces are showing common onnxruntime and libcudart mismatches. This usually happens when Hugging Face updates its base ZeroGPU or Inference Endpoint environment, breaking any Space that doesn’t have strictly pinned dependencies.

While only the authors can ‘fix’ the code to match the new environment, the fact that it’s a multi-platform blackout suggests the platform/upstream dependency changed under them.