sd-webui-controlnet
sd-webui-controlnet copied to clipboard
[Feature Request] Support for batch img2img?
how do we load the same image into the extension as the image loaded into img2img? So we don't have to manually load the image twice each time
Already supported by #36
Already supported by #36
thanks for the reply; What I meant was, not just the img2img
tab, but the batch
subtab of the img2img
tab; when using the batch
tab and the controlnet script, will raise error:
\stable-diffusion-webui\modules\img2img.py", line 76, in process_batch
processed_image.save(os.path.join(output_dir, filename))
AttributeError: 'numpy.ndarray' object has no attribute 'save'
sounds like the output data of controlnet is a ndarray
and when processed as webui's code processed_image
, by line 76, processed_image.save(os.path.join(output_dir, filename))
, will raise an error
yes currently batch only gives you the very first image and does not work for the rest , this shouldnt be that hard to fix
yes currently batch only gives you the very first image and does not work for the rest , this shouldnt be that hard to fix
wonder if we can extract the for loop of the batch process function and output the image into controlnet somehow
Added an option to skip saving controlnet detectmep. May help with this issue
Can confirm this is working now with "Do not append detectmap to output" enabled in settings. Thanks!
Added an option to skip saving controlnet detectmep. May help with this issue
When I use "Do not append detectmap to output" Batch img2img works fine! In my personal opinion, this option would be more convenient to use if moved to the ControlNet panel of the img2img tab.
Added an option to skip saving controlnet detectmep. May help with this issue
great, thanks! just wondering, does this mean that the open pose are no longer guiding the generated images? I noticed that some output images are not following the input image's pose anymore, even with openpose model selected
edit: most still sticks to the input image, but some are not, could just this be noises in latent space?
Found something a little odd. BatchImg2Img runs normally without any problems, but compared to when ControlNet was run in Txt2img (same Seed, prompt), it seems that not only openpose is input, but the image itself is also input. 1: Input Image, 2: Txt2img Openpose, 3: BatchImg2Img Openpose. Is there a way to get the same result as Txt2Img in BatchImg2Img?
I created a temporary extension to fix this issue. By using this extension, img2img processing will not be performed. https://github.com/ddPn08/sd-webui-controlnet-batch-patch
Once #98 is merged it will no longer be needed.
I created a temporary extension to fix this issue. By using this extension, img2img processing will not be performed. https://github.com/ddPn08/sd-webui-controlnet-batch-patch
Once #98 is merged it will no longer be needed.
I double that current outputs are all really dark, which seems to be because it's using the detectmaps as the img2img sources, is this what your PR aims to fix?
@Creative-Ataraxia You can use it like this. For the Patch extension, just enable it. Batch processing works similarly.