This also means I can use Load Clip/device = default instead of "cpu", since the OOM is now handled and tiled vae works.

Perhaps WanImageToVideo should have a tiled option somehow too.

I've update to pytorch 2.7.1+rocm6.3 and this still seems valuable for my gfx1100. It makes sense that all variants of OOM should be handled.

Is there some issue with this approach, or other reason not to merge this?

This is still merge conflict free, want it or shall I close? cc @comfyanonymous @Kosinkadink

Could i get a comment on this, perhaps a short description on why it isn't suitable? Perhaps i could rework it adapt it in some way that still allows oom fallbacks to work on rocm. @comfyanonymous @Kosinkadink

Why don't you set the model_management.OOM_EXCEPTION to whatever the hip oom exception is?

Why don't you set the model_management.OOM_EXCEPTION to whatever the hip oom exception is?

The error this handles is a generic RuntimeError with just the message identifying it as an oom. Matching all RuntimeErrors could include non-oom ones.

RuntimeError: HIP error: out of memory

Test Evidence Check

⚠️ Warning: Test Explanation Missing

If this PR modifies behavior that requires testing, a test explanation is required. PRs lacking applicable test explanations may not be reviewed until added. Please add test explanations to ensure code quality and prevent regressions.

⚠️ Warning: Visual Documentation Missing

If this PR changes user-facing behavior, visual proof (screen recording or screenshot) is required. PRs without applicable visual documentation may not be reviewed until provided.

You can add it by:

• GitHub: Drag & drop media directly into the PR description
• YouTube: Include a link to a short demo