NumPy ABI errors: _ARRAY_API and dtype size changed

Symptoms and scope

These messages usually concern the interface between NumPy and a compiled extension, not a broken workflow connection. Locate the first implicated extension before changing NumPy.

Error fragments for search; IDs, values and filenames may vary:

_ARRAY_API not found
numpy.dtype size changed
A module that was compiled using NumPy 1.x

Source-supported context

NumPy's troubleshooting explains incompatibilities between compiled extensions and NumPy versions. The useful traceback entry is often the first package outside NumPy, rather than the final import wrapper. Source 1 Source 2

Cases to distinguish

1. A compiled extension built for an older NumPy ABI is imported after an upgrade.

2. An old binary remains on the actual runtime's import path.

3. Different custom nodes require incompatible versions of the numerical stack.

Diagnostic sequence

The sequence below is an editorial procedure based on the cited context, not a diagnosis already confirmed for your environment.

Step 1. Record NumPy's version and file location, plus the first compiled package named in the traceback.

Step 2. Check whether that package offers a compatible build. Reinstalling NumPy repeatedly does not rebuild another project's binary.

Step 3. If the failure began with NumPy 2 and no compatible extension is available, consider a documented NumPy 1 fallback only in a recoverable environment and after checking other requirements.

Step 4. Run pip check, import the implicated extension, then execute a small affected task. Preserve the combination that works.

Commands and request examples

Use the environment and shell identified above. Paths and variables are examples to adapt; these commands/snippets were not executed for your machine.

python -c "import numpy; print(numpy.__version__); print(numpy.__file__)"
python -m pip check

Completion check

Both the extension import and its actual operation succeed in the ComfyUI interpreter, with no newly broken declared requirements.

Limitations and cautions

numpy<2 is a conditional fallback, not a universal repair command. One node's workaround can break another node in the same environment.

Original sources

• NumPy troubleshooting import errors — original record checked 2026-09-21.
• pip check — original record checked 2026-09-21.

Localization prepared: 2026-09-22. The source-check dates above were inherited from the existing article; this translation does not claim they were all rechecked today. No GPU run, minimum-memory measurement or runtime guarantee is supplied.

Related troubleshooting and guides

• DLL load failed: identify the native extension instead of downloading loose DLLs
• pip ResolutionImpossible: resolve conflicting requirements, not the download speed