OpenAi has decided to open source their consistency decoder vae
https://openai.com/blog/new-models-and-developer-products-announced-at-devday
Currently no support in Automatic1111, but for ComfyUI use this https://github.com/Jordach/comfy-consistency-vae
Description
initial version
FAQ
Comments (15)
Interesting, I thought you told me not to post models such as this? Is there a clarification on that rule now? Is it allowed for everyone to do or just you?
which model are you talking about? just had a look at your models and the only comment i found was made on the t2i-adapters by h94, where i just asked you to mention the original creator of those models and include a link to the usage instructions. where exactly did i tell you to not post a certain model?
Any plans for an automatic1111 version?
you can already test it on this branch https://github.com/AUTOMATIC1111/stable-diffusion-webui/tree/consistency-dec
@eurotaku Thank you
Please tell me how I can use it.
either use the linked node in comfyui or this branch of automatic1111 https://github.com/AUTOMATIC1111/stable-diffusion-webui/tree/consistency-dec
@eurotakuI can't access the comfyUI node or load it in 1111, is there a solution?
@eurotaku in this new branch I'm having this "MemoryError: bad allocation"
@zaadsatan might simply not be bugfree, yet, still awaiting merge into dev branch. you could ask directly on github, if someone can help you with that error
Anything for SDXL?
So apparently this thing wants to use 38 GB of vram when decoding a 1600x2160 image...
wow, can it be tiled?
bad allocation? why