[bug]: SDXL fails VAE pass when VAE Precision is set to fp16

[radry]

Is there an existing issue for this?

• I have searched the existing issues

OS

Windows

GPU

cuda

VRAM

12GB

What version did you experience this issue on?

3.1.1

What happened?

When generating an image with SDXL and the VAE is set to default with the provided SDXL 1.0 model and the VAE precision is set to fp16, it fails with an error.

Solution: Automatically set VAE Precision to FP32 when VAE is set to Default and model is SDXL based.

Screenshots

No response

Additional context

No response

Contact Details

No response

[psychedelicious]

We can't make the proposed change, because there are sdxl models with fp16 vae baked in.

For now this is necessarily the user's responsibility.

We are working on a model management refactor that should help though.

[radry]

So it can not autodetect which vae it is? That's a problem.

[Millu]

Not as of right now - model management is currently being re-worked, and this should be improved.

[psychedelicious]

You can now select the default VAE precision on a per-model basis.