node to create wildcard label prompt from string list #536
Conversation
modules/impact/util_nodes.py
Outdated
| def doit(self, string_list, prefix_all, postfix_all, restrict_to_tags, exclude_tags): | ||
| # need to access as list due to INPUT_IS_LIST | ||
| # some sanity checks and normalization for later processing | ||
| if prefix_all[0] == None: prefix_all[0] = "" |
There was a problem hiding this comment.
In Python, when checking if something is None, you should use is None.
There was a problem hiding this comment.
Resolved by 43e1bbb
modules/impact/util_nodes.py
Outdated
| }, | ||
| } | ||
|
|
||
| INPUT_IS_LIST = True |
There was a problem hiding this comment.
The behavior seems unclear because string_list is received as a list while the others are not being considered as lists.
It might be better to add a node such as StringListToString instead of directly receiving string_list in this node, and receive input as a concatenated string.
There was a problem hiding this comment.
The review was pending due to my mistake.
There was a problem hiding this comment.
That makes sense.
I have added this node here: e41788a (used WAS node suite WAS_Text_List_to_Text as template).
A new workflow making use of the added node is attached.
There was a problem hiding this comment.
That's awesome
thanks!
I have finally found the time to write up a short tutorial in the impact examples on how I imagine this node to be used, check it out here
The idea behind this node is to automatically create a wildcard prompt for the SEGS detailer. The purpose is to have a single prompt per tile/SEG which guides a denoising process for upscaling an image (in addition to the regular upscale prompt).
This strategy should help to either increase the denoising strength for added details or protect against hallucinations at the same denoise level.
The WD14 tagger of pythongosssss works nicely for assigning tags on the fly for each tile. It is required to use that or something else to create the string list for the node input.
A prototype workflow is attached to this PR.
prompt-per-tile-sample.json