RandomGitUser321

> I think you're using a model from InstantX and that they aren't supported yet. I might be mistaken, but I think only the Flux Controlnets by XLabs-AI are supported...

> I updated it just now an it still uses 10 Gb for a a single lora rank 64: ![image](https://private-user- It likely has to make a complete patched copy of...

I think this is because it's loading the actual transformer/unet when it reaches the ksampler. When it first looks like it's loading the checkpoint, it's probably just loading the t5...

Well blockly is a node based coding system, so a lot of the code is just for the graph of the nodes and all their properties. If you've ever looked...

Personally, I'm not worried about him doing something malicious. You have to think of the diffusion/AI ecosphere though and how a ton of people in it are ultra paranoid individuals(or...

@lllyasviel Thanks for the transparency and explanation. Like I said multiple times, I didn't think you were up to anything shady or anything. I was just kind of speaking about...

I'm trying to figure out something similar as well for a gradio app frontend. Within the websocket portion of my code: ``` def get_images(ws, prompt): prompt_id = queue_prompt(prompt)["prompt_id"] output_images =...

It's likely doing some kind of casting up to float32 or 16 and then back down to fp8, even if you're using an fp8 version of the model. It might...

Yeah I think I was on to something about it upcasting: `supported_inference_dtypes = [torch.bfloat16, torch.float32]` https://github.com/comfyanonymous/ComfyUI/blob/1c08bf35b49879115dedd8ec6bc92d9e8d8fd871/comfy/supported_models.py#L631

> Since upscaling is an important feature, it is very important to remove these artifacts to preserve the original image content. These artifacts are difficult to remove at low denosing...