allow to use del to kill modules (could it be just my keyboard malfunctioning??)
redesign the conditioning area module, or make a new type:
that would allow users to directly draw their separated areas on canvas instead of cauculate all the coordinates. Since all modules are recorded size/position in the json, guess the editor should have all the data needed.
and controlnet/lora should also be used for each separate area. Right now loras can only be linked in a row and only the last one can be linked to the model node in ksampler, so multiple loras in multiple areas is not doable (as far as I have tried)
3.Add more precise controlnet controls: resize the input image, decide which steps to activate controlnet, midas resolution, etc, what webui has. And preprocessor support.
give controlnetapply built-in controlnet loader and image loader, or make a "simple controlnet apply" and the same to t2i networks. That is mainly because those modules can only be linked with the given types in the give way, and once you need to apply a controlnet, it of course will be loaded. Just select it, not create a loader and link it. Also, just click to upload image. Saves lots of time adding loaders. Since controlnet does not have preprocessors atm, no generation output can be used as input in the workflow, so the idea of simple controlnet would not hurt any existing features.
For each image upload module, always use windows file explorer, and each new uploaded file should overwrite the previous in the input folder, that is, dont keep a long image history list in the module, they could never be reused as the list does not have previews for them. And a preview image shoud be shown just like the saveimage module.
allow users to save certain modules and their connections as custom modules
allow editor canvas to auto move with mouse if the user is linking nodes that are located far from each other. ATM the only way to do it is by resizing the whole canvas or moving them together, linking, moving back.
that would allow users to directly draw their separated areas on canvas instead of cauculate all the coordinates. Since all modules are recorded size/position in the json, guess the editor should have all the data needed.
and controlnet/lora should also be used for each separate area. Right now loras can only be linked in a row and only the last one can be linked to the model node in ksampler, so multiple loras in multiple areas is not doable (as far as I have tried)
3.Add more precise controlnet controls: resize the input image, decide which steps to activate controlnet, midas resolution, etc, what webui has. And preprocessor support.
give controlnetapply built-in controlnet loader and image loader, or make a "simple controlnet apply" and the same to t2i networks. That is mainly because those modules can only be linked with the given types in the give way, and once you need to apply a controlnet, it of course will be loaded. Just select it, not create a loader and link it. Also, just click to upload image. Saves lots of time adding loaders. Since controlnet does not have preprocessors atm, no generation output can be used as input in the workflow, so the idea of simple controlnet would not hurt any existing features.
For each image upload module, always use windows file explorer, and each new uploaded file should overwrite the previous in the input folder, that is, dont keep a long image history list in the module, they could never be reused as the list does not have previews for them. And a preview image shoud be shown just like the saveimage module.
allow users to save certain modules and their connections as custom modules
allow editor canvas to auto move with mouse if the user is linking nodes that are located far from each other. ATM the only way to do it is by resizing the whole canvas or moving them together, linking, moving back.