Reference only

Kornia import errors: check the base package, Torch and optional extras

Distinguish a missing base package from an optional feature dependency before changing Torch.

Read the failing name precisely

The current Kornia installation guide lists PyTorch, NumPy and kornia-rs as base dependencies. Its image, onnx and sd extras enable particular features; kornia-rs is not one of those optional extras in the current guide. A base import failure, a named extra failure and an unavailable API symbol need separate checks. The guide's Torch minimum applies to the current release, so compare requirements for the version actually selected by your node. Kornia installation

python -c "import sys; print(sys.executable); import torch, kornia; print(torch.__version__, kornia.__version__)"

Run this with the exact interpreter that launches ComfyUI and retain the traceback if import fails. Compare its package location and versions with the node's declared requirements; a pip show result from another environment does not prove the active import works. Do not apply the newest Kornia/Torch requirement to every older node without checking that node's supported combination.

If the error names kornia_rs, check the base dependency and its native wheel in the active interpreter. If it names an image, onnx or sd extra, verify the task actually uses that feature and preserve the exact missing name. The current lazy-loader can offer to install an extra; an install prompt is not proof that the feature is required by your workflow. If a symbol cannot be imported, compare the node's expected API with the installed Kornia release. Optional extras

After a trusted import succeeds, rerun the original operation with a small image and unchanged settings. Shape, dtype or device errors belong to a runtime branch, not another missing-package attempt. Keep the previous package list and node revision so that a change can be evaluated or reversed without guessing.

Sources and English text reviewed 2026-09-26. No package installation, local import, image operation or GPU reproduction. These are editorial checks, not a diagnosis of your environment.

Didn't solve it?

Check the next possible cause

The same symptom can come from a different cause. Work through these related entries in order.

  1. Spandrel is installed but the model will not load: check architecture supportNo module named spandrelSeparate the library import from architecture support and the identity of the weights.
Paste your full log into the error finder

Was this page helpful?

Anonymous. Only a yes/no counter is stored; no account, IP address or device details.

Sources & references

2026-09-26 checked current official Kornia installation and extras guide; English text reviewed. No package installation, local import, image operation or GPU test.

01Kornia official installation and optional extrasSource checked: 2026-09-26

Report an issue · b940b44b-74d0-559b-97ec-4af1e2b4eedb