Closed Crono141 closed 1 month ago
The upscaling tab uses stable diffusion. Using prompt is a key differentiator between just running a simple single pass upscale model and using SD for upscaling.
You'll need to clear the prompt if you don't want to use it, or use the postprocessing button to use a simple upscaler (much faster btw):
OK, operator error. Thanks.
Is there an existing issue for this problem?
Operating system
Linux
GPU vendor
Nvidia (CUDA)
GPU model
3060
GPU VRAM
12gb
Version number
v4.2.7
Browser
Chrome Latest 128.0.6613.84
Python dependencies
accelerate 0.30.1 compel 2.0.2 cuda 12.1 diffusers 0.27.2 numpy 1.26.4 opencv 4.9.0.80 onnx 1.15.0 pillow 10.4.0 python 3.10.14 torch 2.2.2+cu121 torchvision 0.17.2 transformers 4.41.1 xformers 0.0.25.post1
What happened
After generating, sent image to upscaler. Picked a 4X upscaler without clearing the global prompt first. After upscaling (which takes a long time, longer than generating 4 images), the resultant image had numerous artifacts affected by the prompt
What you expected to happen
I expect the upscale to only upscale, and not also take prompt and use it to fill in the upscaling
How to reproduce the problem
Generate an image using invokeai. Tweak in the unified canvas. Save to library. Right click send to upscaler. Scale by 4X without clearing global prompt. See if anything ugly appears
Additional context
The issue goes away, using the exact same settings for the upscaler, if the global positive prompt is cleared before invoking.
Discord username
No response