Your Portrait + Any Background: ACE Plus & Redux Make Them One in ComfyUI

Introduction to Advanced Scene Integration

Blending Reality and Imagination in ComfyUI

Imagine dropping your portrait subject into any environment – whether a tropical beach at sunset or a cyberpunk cityscape – while preserving their identity and creating seamless interactions with the scene.

This article introduces a game-changing workflow that transforms basic background swaps into context-aware integrations, where subjects appear to belong in their new surroundings.

The Evolution from Basic Swaps

Building on our previous ACE Plus portrait work (where we achieved 99% face similarity), this workflow solves three critical challenges:

1. Environmental Sync: Matching subject lighting/shadows to the new scene
2. Contextual Interaction: Creating natural contact points (e.g., feet in water)
3. Identity Preservation: Maintaining facial features and clothing details

Ready to turn disjointed elements into cohesive visual stories? Let’s first understand the workflow’s two-stage philosophy before diving into the node groups.

Workflow Overview: Two-Stage Transformation Process

Uploaded Portrait:

Uploaded background:

12 Groups? Only 5 Are Essential!

Don’t let the node count scare you – by Group 5, your subject is naturally placed in the scene. The remaining groups are optional tweaks for perfectionists.

Core Setup (Groups 1-5)

• Positions subject perfectly
• Matches lighting/shadows
• Fixes obvious issues (e.g., awkward limbs)
• Restores details like facial clarity
• Result: A “ready-to-post” integration

Optional Polish (Groups 6-12)

• Groups 6-9: Redux tweaks textures/poses + ACE face locking
• Groups 10-12: Fixes clothing glitches & hands
• Pro Tip: “Only use these if flaws bother you!”

Ready to begin? Let’s start with Node Group 1.

Support me on Patreon and Get this workflow: https://www.patreon.com/posts/exclusive-and-124319336

Video Tutorial:

Node Group 1: Subject Positioning & Depth Blur

Where Your Workflow Begins

Goal: Place your subject in the scene with natural background blending.

3 Key Steps

1. Clean Workspace Setup
   • Disable all groups except 1 & 2
   • Upload portrait + background images
2. Automatic Depth Blur
   • The ProPostDepthMapBlur node (your secret weapon) creates background bokeh based on spatial relationships
   • Result: Instant depth realism without manual masking
   
3. Surgical Edits
   • Use the Preview Bridge node’s magic eraser:
     • Delete problematic areas (e.g., forearm section)
     • Watch background elements peek through (like flowers in our demo)

Node Group 2: Powerhouse Model Loading & ACE Plus Setup

Your Workflow’s Engine Room

Goal: Load essential tools fast while prepping for face/clothing swaps.

3 Key Components

1. TeaCache Turbo Boost
   • Caches frequent operations like a racing pit crew
   • Result: faster rendering vs standard loading
   
2. Florence2Run Auto-Pilot
   • Generates scene-specific prompts automatically
   
3. Dual ACE Plus LoRAs
   • Top LoRA: Face swap specialist (preserves identity)
   • Bottom LoRA: Outfit change master (my testing shows superior clothing edits)

Why This Combo Rocks

• TeaCache handles speed → You focus on creative choices
• Florence2Run eliminates prompt-guessing → Faster iterations
• Dual LoRAs prevent “jack-of-all-trades” model compromises

Node Group 3: Dynamic Lighting Overhaul

Where Your Scene Gets Its Mood

Goal: Transform flat lighting into dramatic environmental ambiance using IC-Light.

The Magic & The Reality

1. Run the Workflow →
   • Instantly adds directional lighting (e.g., backlight mimicking sunset windows)
2. Expected Trade-Offs →
   • Details get fuzzy (we fix this in Group 4)
   • Colors temporarily dull (part of the process)

Node Group 4: Detail Resurrection Squad

Fixing IC-Light’s Fuzzy Side Effects

Goal: Recover crisp details while preserving IC-Light’s dramatic lighting.

3-Step Cleanup Process

1. Contrast Rescue Mission
   • Salvages lighting details IC-Light muted
   
2. Background Time Machine
   • Reintroduces original background elements
   
3. Secret Sauce: Detail Transfer
   • Image Detail Transfer Node:
     • Start at blur_sigma=5
     • Bump to higher if artifacts exist
   

Node Group 5: Shadow Sculpting & Environmental Anchoring

Where Your Subject ‘Lands’ in the Scene

Goal: Fix awkward details and generate realistic shadows that ground your subject.

3-Step Shadow Crafting

1. Fast Bypasser Prep
   • Temporarily disable KSampler
   • Run workflow
   
2. Magic Marker Masking
   • Paint problem areas (e.g., jagged arm edges)
   
3. Detail Generation
   • Reactivate KSampler → Automatic detail rendering

Real-World Results

• Arm Edge Fix: Blends forearm seamlessly into floral background
• Floor Shadow: Adds weight to feet (no “floating” effect)
• Hair Highlights: Backlight wraps naturally around strands

Node Group 6: Redux Repainting Revolution

Precision Control for Scene Reinvention

Two Critical Choices:

1. Redux Reference Image Selection

• Option A: Group 1 Output (Original Layout)
  • Use Case: Reset to initial positioning
  • Risk: Loses shadow/lighting work from Groups 3-5
• Option B: Group 5 Output (Shadow-Enhanced Image)
  • Transcript Directive: “Always pick Option B” for realistic lighting continuity

2. KSampler Latent Image Options

• Path 1: Empty Latent Image
  • Denoising Strength: 1.0 (Total reinvention)
  • Effect: Drastic pose/background changes (e.g., sitting → standing)
• Path 2: Group 5 Image
  • Denoising Strength: 0.3-0.7 (Controlled tweaks)
  • Effect: Preserves composition while refining textures/edges

Workflow Walkthrough

1. Test Path 1 (Empty Latent):
   • Set denoising=1 → Run workflow
   • Result: Radical transformation (e.g., beach → neon cityscape)
2. Switch to Path 2 (Group 5 Image):
   • Set denoising=0.5 → Rerun
   • Result: Keeps 90% of prior work but:
     • Fixes lighting gradients
     • Sharpens eroded details
     • Trade-off: Face distortions occur (fixed in Group 7)

Node Group 7: Facial Fidelity Rescue Squad

Re-Anchoring Faces to Source Likeness

Goal: Fix Redux-induced facial distortions while preserving environmental lighting.

Node Group 8: Face Fusion Control Center

Seamless Original/Generated Face Hybrids

Goal: Merge the source portrait’s face with Redux-adjusted features without seams or style breaks.

3-Step Merge Protocol

1. Surgical Stitching
   • Combines:
     • Original face (Group 1)
     • Redux-adjusted face (Group 7)
2. Precision Mask Growing
   • MaskGrow Node: Pro Tip: “Expand mask 20% beyond face edges – but don’t swallow the neck!”

Node Group 9: ACE-Powered Identity Lock

Final Face Swap with Environmental Harmony

Goal: Swap faces while maintaining 99% source likeness and scene-appropriate lighting.

3-Step ACE Protocol

1. Fill Model Activation
   • Paired with ACE Portrait LoRA
   • Default: denoising_strength=0.8 (ideal lighting balance)
2. Nuclear Option
   • Set denoising_strength=1 when:
     • Facial features drift (＜95% similarity)
     • Eyes/nose lose definition
3. Seed Roulette
   • Hit “Manual Random Seed” 3-5 times → higher success rate in trials

Node Group 10: SAM-Powered Garment Extraction

Precision Clothing Isolation for Style Integrity

Goal: Create pixel-perfect clothing masks while retaining original garment essence.

3-Step Garment Capture

1. Outfit Verbal Blueprint
   • Input detailed description:
     • “Red cotton t-shirt with ‘Vintage’ white text”
     • “Distressed jeans, knee rips, 90s wash”
   • Critical: Mention logos/textures → SAM uses these as anchors
2. SAM-Powered Mask Generation
   • Segment Anything Modelnodes:
3. Stitching
   • Stitching clothing images using IC Mask node:

Node Group 11: Clothing Transplant & Final Match Check

Where Repainted Clothes Meet Original Precision

Goal: Blend repainted clothing seamlessly using the same crop/paste technique from face swaps.

3 Simple Steps

1. Run Workflow →
   • Generates repainted clothing based on Group 10’s mask
2. Crop & Paste:
   • Isolate repainted area → Paste onto original image
   • Uses same method as Group 9’s face swap
3. Compare Results:
   • Toggle comparison node → See near-perfect clothing match

Node Group 12: Fix Hands

Optional hand refinement for images requiring detailed appendage correction.

Conclusion

This workflow transforms intimidating scene integrations into a logical, two-stage journey. Whether you prioritize speed (stopping at Group 5) or perfection (unleashing Groups 6-12), you’ll maintain face/clothing authenticity while achieving seamless environmental harmony. Follow for more ACE Plus deep dives – next week, we’re reinventing portrait lighting without a single camera!