Update Highlights – Version 1.7

Prompt system rebuilt (11 segments, 4 chunks)

• The positive prompt is now split into 11 dedicated boxes, each holding one idea: Quality, Style, Character, Body, Outfit, Expression, Pose, Partner/Interaction, Camera, Scene, Lighting.

• The boxes are merged in a logical Danbooru order, in 4 chunks separated by BREAK: Quality + Style BREAK Character + Body + Outfit BREAK Expression + Pose + Partner BREAK Camera + Scene + Lighting

• Empty boxes are skipped automatically: no stray commas, no empty chunks.

• A live preview node shows the final merged prompt.

Wildcard preset library (PI)

• New library of 68 presets in custom_wildcards/PI/<category>/<preset>.txt.

• Categories: chars, body, outfit, expression, pose, pose_nsfw, partner, camera, scene, lighting, style.

• Switching character, pose or outfit now means changing one line.

Lock Panel

• Each prompt box has its own seed, grouped in one panel.

• Lock any segment (e.g. keep the pose, randomize the outfit) with Use last queued seed; unlock with Randomize Each Time.

• The same seed always gives the same wildcard result, so a locked segment is fully reproducible.

Negative prompt split in two

• Base negative: quality and anatomy terms, always active.

• Scene negative: a small box for scene-specific terms (unwanted objects, people, body parts) that you change per image.

Detailer: anatomy fix, same look

• The full-image Detailer keeps the original high-detail look (full prompt, same settings as before).

• Added Kohya Deep Shrink (PatchModelAddDownscale): the model builds shapes at its native resolution before refining at 2× size. This prevents duplicated anatomy (double navels, extra body lines) at high resolution.

• Added a light Tile ControlNet (Union ProMax, strength 0.35) as an anatomy guard. Set its strength to 0 to disable it.

Fast Preview mode

• One switch in the control center skips the Hi-Res Fix and every Detailer.

• Find the right composition in seconds, then switch it off for the final render.

• Preview files are saved with a PREVIEW_ prefix so they never mix with final renders.

Color and exposure fix

• The final color-harmony pass was too strong and washed out renders (lifted blacks, lower contrast, softer details).

• Its strength is now 0.3: colors stay consistent, while blacks and sharpness are preserved.

Other changes

• The (realistic:1.2) tag moved from the quality block into the Style box, so you can change the rendering style without touching the quality tags.

• Typo fixes in the default prompts.

• Node titles and groups cleaned up.

Update Highlights – Version 1.6

• Completely redesigned to be more user-friendly, compact, and efficient, making the workflow easier to navigate while preserving its full flexibility.

• The Dual Sampler and Hi-Res Fix pipeline have been reworked from the ground up, resulting in better image quality, improved coherence, and a more reliable generation process.

• The entire Detailer section has been rebuilt with optimized models and a cleaner layout, providing more consistent face and hand refinement while remaining easier to configure.

• The ControlNet section has also been fully revised, improving compatibility, organization, and overall usability for reference-based generations

Update Highlights – Version 1.5

• Completely redesigned for better readability and ease of use, heavily inspired by vslinx’s workflow (shoutout to him).

• The preview section has been revised for clearer and more intuitive feedback during generation.

• The old dual-sampler setup has been replaced with a dedicated Hi-Res Fix node, improving image coherence and reducing anomalies during upscaling.

• A new image filter now generates four base images, allowing users to select the most suitable one before moving forward with detailing.

Update Highlights – Version 1.4

• Introduced a dedicated second sampler integrated with a Highres Fix script, significantly enhancing image fidelity and detail sharpness in final outputs. Expect smoother upscale transitions and richer textures, especially for close-ups or larger compositions.

• Seamless integration of the new highres logic into the existing pipeline — no extra setup needed. Just plug and play for better results.

• Further minor improvements to node connections and internal logic to ensure a more stable and optimized generation flow.

Update Highlights – Version 1.3

• Added support for wildcards, allowing dynamic and flexible prompt generation

• Introduced the ability to randomize wildcards for varied and automatic prompt creation across generations

• Improved overall user experience with better layout clarity and workflow accessibility

• Workflow remains modular and optimized for structured Multi-LoRA character and concept building

Update Highlights – Version 1.2

• Refactored the "Adetailer" tabs for improved readability and usability

• Added a control panel to easily toggle ON/OFF for various node groups

• Globally improved layout structure and visual clarity for smoother navigation

• Replaced and optimized samplers used within the "Adetailer" sections

Update Highlights – Version 1.1

Implemented a dedicated section for nipple detailing, with a toggleable node allowing users to enable or bypass the detail processing of each detailler section as needed.

Quick Start

1. Install the custom nodes and models listed below.

2. Copy the PI wildcard folder into ComfyUI/custom_nodes/comfyui-impact-pack/custom_wildcards/.

3. Load the workflow, choose your checkpoint and LoRAs in Base and Prompting.

4. Fill the prompt boxes (or use presets, see below).

5. Turn on ⚡ Fast Preview, queue, and pick your favourite of the 4 images in the Image Filter.

6. Lock the segments you like, turn Fast Preview off, and render the final image.

Key Features:

Structured Multi-LoRA framework

• LoRAs organized in one LoRA Manager loader, with trigger words toggled on and off.

• Clear visual hierarchy for easy navigation.

Precision enhancement tools

• Separate Detailers for face, eyes, hands, body and NSFW, each toggleable from the control center.

• Built-in ControlNet for image-based generation (depth, pose, tile…).

• Full support for embeddings and A1111 prompt weighting.

Optimized Illustrious foundation

• Pre-configured for stable, high-detail Illustrious renders.

• Modular design: every aspect (character, pose, scene, lighting) is its own block.

Fast iteration

• Fast Preview, 4-image selection, and the Lock Panel take you from concept to final render quickly.

• Locked segments keep results consistent across generations.

Perfect for

• Character designers

• Concept artists who need fast iteration

• Style explorers who want maximum control

Download includes

• Main workflow JSON

• PI wildcard preset library

Ressources :

Detailers Models :

• Face: Mask aDetailer (eyes, eyebrows, nose), ADetailer Anime Girl Face

• Eyes: PitEyeDetailer v2

• Lips: ADetailer Lips Model

• Nose: ADetailer – Nose Model

• Hands: hand_yolov9c.pt

• Nipples: Anime Girls Nipple Adetailer, ADetailer – Nipples Model

• Vagina: Pussy Adetailer

• Penis: Cock & Ball Detection

Nodes used:

• ComfyUI Impact Pack

• ComfyUI Impact Subpack

• rgthree's ComfyUI Nodes

• ComfyUI-Crystools (new in V11: Fast Preview switches)

• WAS Node Suite (prompt merging)

• ComfyUI-KJNodes (color match)

• ComfyUI LoRA Manager

• efficiency-nodes-comfyui

• ComfyUI Image Saver

• ComfyUI_Comfyroll_CustomNodes

• ComfyUI-Easy-Use

• ComfyUI-Custom-Scripts

• ComfyUI_Mira

• WLSH Nodes

• Masquerade Nodes

• SD Prompt Reader

• cg-image-filter

Join the Discord!

Come hang out, share your AI art, give feedback, ask questions, or just vibe with the community.
It’s also the best place for troubleshooting, commissions, and sneak peeks of upcoming work.

👉 https://discord.gg/zBK8QNZ7xt