Drift
8/12/2025, 2:56:53 AM
No.938334837
>>938334313
no worries.
so, it might be hard to follow since you're just beginning, but here's the main idea:
get was-node-suite-comfyui extension, it's very useful beyond just this workflow.
- learn to use the 'image generate gradient node' in the horizontal direction. this node is from that extension and the gradient is used to create a rectangular area, all other nodes will be from core comfy
- create three regions for your characters with those. they should be close to disjoint, but a very small overlap is fine
- 'convert image to mask' on the outputs. the 'channel' doesn't matter.
- use the masks for the 'condition (set mask)' nodes with weight 1 and 'mask bounds' setting. the conditioning on each comes from the 'clip text encode' nodes, which are your positive prompts.
- have one 'solidmask' which is the main prompt and it also leads into 'condition (set mask)', but with strength 0.5
- the outputs of all those set masks nodes go into 'conditioning (concat)' nodes, two by two, and then to the KSampler.
that's basically it. a more advanced way would be using conditioning (combine), but that's way more advanced for now
no worries.
so, it might be hard to follow since you're just beginning, but here's the main idea:
get was-node-suite-comfyui extension, it's very useful beyond just this workflow.
- learn to use the 'image generate gradient node' in the horizontal direction. this node is from that extension and the gradient is used to create a rectangular area, all other nodes will be from core comfy
- create three regions for your characters with those. they should be close to disjoint, but a very small overlap is fine
- 'convert image to mask' on the outputs. the 'channel' doesn't matter.
- use the masks for the 'condition (set mask)' nodes with weight 1 and 'mask bounds' setting. the conditioning on each comes from the 'clip text encode' nodes, which are your positive prompts.
- have one 'solidmask' which is the main prompt and it also leads into 'condition (set mask)', but with strength 0.5
- the outputs of all those set masks nodes go into 'conditioning (concat)' nodes, two by two, and then to the KSampler.
that's basically it. a more advanced way would be using conditioning (combine), but that's way more advanced for now