T shapes cannot be multiplied.

[modmodmodmod]

Hi, When using ControlNet, the following error occurs.

Error occurred when executing KSampler: linear(): input and weight.T shapes cannot be multiplied (77x2048 and 768x320)

There is no problem if omit yellow group in the image. The error is occurring only within ControlNet. The same thing happens with Openpose.

What does it mean by "T shapes" ?

environment: Apple M1/MacOS 12.7.4 StabilityMatrix 2.9.2 ComfyUI latest version Python 3.12.2 pip(3) 24.0

[modmodmodmod]

This error no longer occurs when changing the Checkpoint model to something other than the base models. Is the combination of the SD base model and ControlNet bad?

[NeedsMoar]

Probably. Controlnet might make assumptions that only apply to models trained with the koyha-ss scripts (new weights they insert maybe?) since pretty much nobody runs the base model. If nothing else a good reason to run some other newer general purpose model is that most were trained against larger images than the originals. Models closer to the base model tend to produce terrible images at resolutions above 512x512 where the newest of the SD 1.5 models don't have issues generating 1024x1024 or non-square resolutions in that neighborhood directly without the koyha downscale node inserted.

I can't say for sure because I don't even have a copy of SD 1.5 (or SD 2.1) on my computer... I can't remember the last time I actually ran one of them.

[modmodmodmod]

Thank you so much!