PreSampling (LayerDiffuse ADDTL)

This is the companion node to easy preSamplingLayerDiffusion, and it only earns its place in your graph if you set that node's method to Foreground or Background. Everything else in the LayerDiffusion pipeline works off one shared prompt from your pipe - this node is what lets you give the foreground layer, the background layer, and the blended composite each their own separate prompt, instead of forcing one description to cover both.

Why layers need separate prompts. LayerDiffusion in Foreground/Background mode isn't generating one image with a transparent cutout - it's generating two things that are meant to composite together (say, a subject on one layer and an environment on the other), and each half of that composite benefits from its own description. "A woman in a red dress" describes the foreground; "a rainy city street at night" describes the background; and a third description of how they should read together once combined is the blended prompt. Trying to cram all three into a single prompt on the base node gets you a muddled result where the model can't tell which part of the description applies to which layer.

Inputs. foreground_prompt, background_prompt, and blended_prompt - all required, multiline text boxes, default empty. There's also an optional path around text entirely: optional_fg_cond, optional_bg_cond, and optional_blended_cond accept pre-made CONDITIONING directly, for when you've already encoded a prompt elsewhere (with custom weighting, say, via easy pipeEdit) and want to feed that in rather than retype it here. Required pipe in, pipe out - same pattern as every other presampling node in the pack, chain it before your easy kSampler.

Install. Same base pack as always:

cd ComfyUI/custom_nodes
git clone https://github.com/yolain/ComfyUI-Easy-Use

or through ComfyUI Manager, search "ComfyUI Easy Use," then restart. Because this node only matters in combination with LayerDiffusion, the same extra step applies here too: the pack's README notes that LayerDiffusion features may need pip install -r requirements.txt run once inside the cloned folder before anything works.

Common issues. The node itself is simple enough that most trouble actually traces back to its partner node rather than to this one - if easy preSamplingLayerDiffusion's method is set to Everything or one of the Injection modes rather than Foreground/Background, this node's prompts are effectively ignored, because the base node isn't generating separate layers to apply them to. Double-check the method setting first if your foreground/background prompts don't seem to be doing anything. And remember the whole LayerDiffusion pipeline this feeds into is SD1.5/SDXL-only with tooling that's been dormant since 2024-2025 - if you're not already committed to that model family for this shot, this node isn't the piece worth troubleshooting around.