Open andupotorac opened 2 months ago
I wanted to know how we can tell what version of compel is used in this, and if we can bring up to date somewhere. In their examples the apricots drop significantly with several -.
Most of the weight interpretations are very similar, unless you are using specifically weighted prompts.
a (man:1.2), picking (apricots:0.7) from a tree
Most of the weight interpretations are very similar, unless you are using specifically weighted prompts.
a (man:1.2), picking (apricots:0.7) from a tree
Note I've selected the Compel prompt syntax, which uses + and - for weighting.
Not in this context. The weighting in comfy (or at least BNK specifically) is done through the (text:weight) prompt syntax, regardless of which weight_interpretation is selected
Not in this context. The weighting in comfy (or at least BNK specifically) is done through the (text:weight) prompt syntax, regardless of which weight_interpretation is selected
Oh, is this anywhere in the docs? Because otherwise why would it even list Compel? I tried Blend for example and it works. Plus Compel uses this format: (apricots)1.1
@BlenderNeko can you confirm please?
@BlenderNeko I did some tests. In Compel they suggest using values between 0 and 2, 1 being the default. But when I use 2 (for many flowers), the image breaks.
What would be the min and max values to be used in your implementation of Compel?
I tried Blend and Conjunction and I saw that those things do work.
I also tried the +++ and the (word)weight syntax that compel is using, and it doesn't seem to work.
I tried (flowers:1.5) and this did work - see the flowers in the background.
I also tried to remove the flowers using (flowers:0) but that also breaks the image. (same image broken like the one above). If I use 0.1 for the weight (flowers:0.1) it does remove them. So maybe this is the min limit?
Can you share what is and isn't similar to Compel please, so we can use your node the right way? Thanks!
One final question, related to Batch input. I see it's in your code, but how should we call it from the Node so that it works?
Eg. how would we call this?
["(photorealism:1.5) cute samoyeed, lying on the grass, relaxed expression, surrounded by (flowers:0.1), bamboo fence, sunny day, detailed fur, high detail, photorealistic, vibrant colors", "a black cat on a bench"]
I tried and it doesn't work. It mixes the results into one photo.
I've used this workflow but the number of apricots is not low.