ComfyUI-QwenEditPlusSlots

ComfyUI-QwenEditPlusSlots

ComfyUI custom nodes for Qwen Image Edit workflows:

• Text Encode Qwen Edit Plus (Slots) — assembles a prompt from up to 10 toggleable, reorderable text slots (with a live, editable preview) and encodes it through the Qwen Image Edit Plus pipeline, outputting CONDITIONING plus the final prompt as STRING. A drop-in replacement for the core TextEncodeQwenImageEditPlus node with a built-in prompt builder.
• Skin Realism (De-Plastic) — post-processing that fixes the plastic, waxy skin Qwen (and diffusion models generally) often render on people.

Install

Via ComfyUI-Manager: Install via Git URL → paste this repo's URL, then restart.

Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/normbarrette-arch/ComfyUI-QwenEditPlusSlots.git

Restart ComfyUI (hard-refresh the browser too, so the slot UI JavaScript loads). The node appears under advanced/conditioning as Text Encode Qwen Edit Plus (Slots).

Node: Text Encode Qwen Edit Plus (Slots)

Inputs

• clip (CLIP) — required.
• vae (VAE), image1..image4 (IMAGE) — optional Qwen reference images. The core node stops at 3; image4 is an opt-in extra (the model is optimized for ≤3, so a 4th may reduce consistency). Unused image inputs are skipped, so an unconnected image4 behaves exactly like the 3-image core node.
• mode — Combine Enabled (join enabled, non-empty slots with separator) or Select One (output a single chosen slot).
• select, separator, slot_count (1–10) — combine/select controls and how many slots are visible.
• use_manual_edit — AUTO: encode the combined slots; MANUAL: encode the hand-edited final_prompt text instead.
• final_prompt — live preview of the combined slots while AUTO (read-only); becomes a free, encoded-verbatim edit field while MANUAL.
• text_1..10 + enable_1..10 — the slots, each with a native ON/OFF toggle and ▲ up / ▼ down reorder buttons. Order persists with the workflow.

Outputs

• conditioning (CONDITIONING) — identical to what core TextEncodeQwenImageEditPlus produces for the same final prompt + images.
• prompt (STRING) — the final text that was encoded (assembled or manual).

Credits & licensing

• The encode pipeline mirrors ComfyUI core's TextEncodeQwenImageEditPlus (comfy_extras/nodes_qwen.py), which is GPL-3.0; this project is therefore licensed GPL-3.0 — see LICENSE.
• The slot UI JavaScript is adapted from kymeraj/comfyui-prompt-builder (MIT) — see THIRD-PARTY-NOTICES.md.

Node: Skin Realism (De-Plastic)

image/postprocessing category. Fixes plastic-looking AI-rendered skin by synthesizing what the render lacks — no reference photo needed:

• pores — band-passed micro-texture, midtone-weighted, automatically attenuated where the image already has texture (never double-textures).
• grain — fine luminance grain that unifies the treated region.
• sheen_reduction — soft-knee compression of skin highlights (kills the waxy specular look).
• mottling — very low-frequency color variation (subsurface redness).

Targeting: by default it detects skin-colored regions (approximate, color-based, cleaned + feathered). Wire any MASK into mask (e.g. from a person/face segmenter) to override precisely. detect_skin OFF + no mask = treat the whole frame.

Inputs: image, master strength, texture_scale (pore size; ~1.0 for a face filling a ~1024px frame), the four effect sliders, detect_skin, seed (deterministic texture), optional mask.

Outputs: image, plus treated_mask so you can inspect exactly what was touched.

Pure torch + numpy — no extra dependencies.