Replies: 5 comments 6 replies
-
Beta Was this translation helpful? Give feedback.
-
I'm having the same issue. In a1111 my workflow consisted of rendering a decent image and then loading it into the ControlNet (scribble) as a "reference image" (not sure of the terminology). Now I could scratch in a few specific details in any area of the photo and like magic, I could create the exact details I intended. But, yeah... In Forge it doesn't seem to work at all. I can still load an image, but tracing or sketching in new details has zero effect. :( |
Beta Was this translation helpful? Give feedback.
-
That's your mistake, not ControlNet's :) |
Beta Was this translation helpful? Give feedback.
-
Before responding that someone is incorrect, it might be beneficial to review the question more thoroughly. Are you referring to Forge, or Auto1111? According to the original poster (OP), the 'sketch (in place)' function seems to be missing from the Forge controlnet implementation, and my observations confirm this. The option to select 'none' in the 'preprocessor' field does not appear to resolve this issue, nor does any setting that allows for sketch input. It would be more constructive to focus on understanding and addressing the actual issue at hand rather than seeking to correct others. If the solution is unclear, perhaps further investigation or assistance from others could be useful. |
Beta Was this translation helpful? Give feedback.
-
Just a bit off-topic... You mentioned Photopea. I was unaware of that until now. You say it uses an internet connection to send data to their service? But I'm reading otherwise online. Any clarification around this? Either way, it looks pretty cool. |
Beta Was this translation helpful? Give feedback.
-
I'll start this as a discussion instead of a bug report since I don't know if this is the intended behavior or not;
It seems that in forge with the integrated Controlnet both the Sketch and Scribble preprocessors aren't working the same way they do in regular a1111 webui in that they only seem to accept uploaded images and not support drawing manually on the canvas neither by creating a blank canvas and drawing on it or uploading an image and drawing on top of the image.
Is anyone else experiencing this?
For example
Using cyberrealistic_v41BackToBasics checkpoint with prompt
a large grassy field, a house in the center of a grassy field, realistic
Neg:art, 3d, cartoon
and seed 4272455667 with the following Controlnet scribble:The image produced is different than without controlnet enabled but doesn't at all resemble the shape scribbled:
With Controlnet:
Without:
The scribbled image is shown as a separate image though in the output so it's at least being intercepted from the Controlnet input it seems;
Beta Was this translation helpful? Give feedback.
All reactions