You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
This gets to be a a problem when one of the selected checkpoints doesn't have a baked VAE, etc. The VAE from the "Efficiency Loader" that ends up in the "SCRIPT" output gets used regardless of what's fed into the KSampler (Efficient) input, though one might expect that connecting the optional input would be the overriding VAE.
In the example image/workflow attached, If you load an SDXL model in the plot with an SD 1.5 VAE (to make it obvious where it's using the wrong vae), and select the SDXL VAE in the loader connected to the "optional_vae" input, it demonstrates what's going on. The latter uses a VAE Decode node instead - it works, but if you don't let the KSampler do the decoding, you don't get the XY Grid.
Disconnecting the "Dependencies" wire from Loader to XY Plot node fixes the VAE issue, but it also keeps the grid from generating, just like bypassing the decode on the KSampler, the plot script seems to be the culprit.
Update: Looking at the code (and confirming in the workflow) this morning - even more strangely, connecting an "optional_vae" is required when doing an XY plot.
The text was updated successfully, but these errors were encountered:
This gets to be a a problem when one of the selected checkpoints doesn't have a baked VAE, etc. The VAE from the "Efficiency Loader" that ends up in the "SCRIPT" output gets used regardless of what's fed into the KSampler (Efficient) input, though one might expect that connecting the optional input would be the overriding VAE.
In the example image/workflow attached, If you load an SDXL model in the plot with an SD 1.5 VAE (to make it obvious where it's using the wrong vae), and select the SDXL VAE in the loader connected to the "optional_vae" input, it demonstrates what's going on. The latter uses a VAE Decode node instead - it works, but if you don't let the KSampler do the decoding, you don't get the XY Grid.
Disconnecting the "Dependencies" wire from Loader to XY Plot node fixes the VAE issue, but it also keeps the grid from generating, just like bypassing the decode on the KSampler, the plot script seems to be the culprit.
Update: Looking at the code (and confirming in the workflow) this morning - even more strangely, connecting an "optional_vae" is required when doing an XY plot.
The text was updated successfully, but these errors were encountered: