Closed rugabunda closed 9 months ago
What were your generation settings when you experienced this error? If you could share the model you were using that would be very helpful!
What were your generation settings when you experienced this error? If you could share the model you were using that would be very helpful!
default... if I remember correctly; it was a completely clean install; I believe I was using an imported SDXL and SDXL refiner... I tried both and just SDXL. Just updating to RC2 before I read this comment, so I will see what happens after its complete
Same problem....
default fresh install rc2
default generation settings, using imported sd_xl_base_1.0_0.9vae exclusively, 1 it, 50 step, 7.5 cfg, Euler, vae default, fp16/32
changing generation settings does not seem to change the error
Ok, solution: using stable-diffusion-xl-base-1-0 works, the imported sd_xl_base_1.0_0.9vae does not work.
I notice invoke downloads 12 gigs worth of sdxl model; including the tokenizer and text incoder; whereas a1111 only needs the base file; why is that? and is this why the vae version I downloaded does not work?
I see, when I imported the vae version, i had selected the wrong config file... i was using inference rather than sd_xl_base.yaml
changing that solved this problem.
curious though, why invoke downloads the tokenizer, text encoder (etc), what do they do? whereas most checkpoints are just a single file. is there a benefit to using invokes sdxl model over a single sdxl base checkpoint?
Is there an existing issue for this?
OS
Windows
GPU
cuda
VRAM
12
What version did you experience this issue on?
3.5.0 RC1
What happened?
Fresh install... loaded without error, received the following error upon invoking image:
Screenshots
No response
Additional context
Full log:
Contact Details
No response