PSD Layer Compositor

The splitter took your PSD apart; this node puts it back together - with one crucial difference: it lets you swap a layer (or a whole run of layers) for an image you generated, then recomposites everything at the original canvas size using the manifest's positions, opacity, and visibility. It's the workhorse of the "AI background replacement in a real design file" workflow, and the version in the README has grown three replacement modes precisely because real PSDs are messy.

The three replacement modes

• single - replace one layer at replacement_index (0 = bottom, usually the background) with replacement_image. The original behavior, and the one that covers a clean PSD.
• replace_range - swap a contiguous range [replacement_index..replacement_end_index] with one image. This is for the messy case where the "background" is actually five stacked layers (gradient, texture, soft shadow, sky, wall) and no single layer is the background. range_fit decides whether the replacement is sized to the full canvas or to the union bounding box of the replaced layers.
• underlay - paste the replacement under all original layers. It's the escape hatch when you genuinely can't pin down which layers are the background; you get the new backdrop behind everything and toggle offending layers' visibility off in the source PSD.

The README's ladder is worth internalizing: try single → detect a range (with PSDBackgroundDetect feeding the indices) → fall back to underlay. Each mode trades automation for reliability.

replacement_fit (stretch / fit / cover / center) and background_color (transparent / white / black / hex) round out the basics - fit letterboxes and cover crops, so pick the one that doesn't distort your generated art.

The fancy part: text recolor

Behind the optional inputs lives real polish: text_recolor_mode can recolor text layers in the output. manual uses text_color for every text layer; auto_contrast picks black or white per layer from the luminance behind it; complement follows the magazine convention (white on dark, black on light). Combined with text_shadow, this keeps captions readable when you've dropped in a wild new background. When output_psd_path is set, recolored text is saved as a clipping PixelLayer above each TypeLayer - so the text stays editable in Photoshop.

The outputs

image is the flat recomposite at original canvas size - the thing you actually wanted. change_log is a detailed run-down of what changed (replaced layers, recolor decisions per layer, shadows, skipped layers, output path); summary is the one-sentence version; psd_path is where the re-exported PSD landed, if you set output_psd_path. That save works in single mode only in v1 - a documented limitation, so don't expect a re-exported PSD from a range swap yet.

Installing it

Part of TrentNodes; install the pack once:

cd ComfyUI/custom_nodes
git clone https://github.com/TrentHunter82/TrentNodes.git
cd TrentNodes
pip install -r requirements.txt

ComfyUI Manager ("Trent Nodes") works too, modulo the known day-one repo-rename quirk that makes Manager sometimes flag the pack as "unsafe" - clone manually if it refuses. Needs psd-tools (in requirements) and the splitter's output folder; no model downloads.

Where people get burned

• replacement_index semantics - it's the bottom layer that's index 0, matching the splitter. A background that's near the top of the layer panel will not be index 0.
• Range indices - replacement_end_index is inclusive. Off-by-ones here are the most common PSD complaint.
• Hidden layers reappearing - respect_visibility (default on) skips them; flip it off only if you deliberately want hidden layers composited.