<h1 id="modular-multi-pass-refinement-workflow">✨ Modular Multi-Pass Refinement Workflow</h1>This workflow is built for clarity, flexibility, and iteration. The entire graph is laid out cleanly with all lines routed visibly and color-coded, so it’s easy to follow without spaghetti connections. Explanatory notes are placed throughout to guide you, and five hotkeys give quick access to the key panels: 1 = Control Panel, 2 = Gallery, 3 = Positive Prompt, 4 = Post Processing, 5 =Docs. From lightweight Pass 1 tests to full multi-pass refinements with face swaps, detailers, expression editing, and labeling, every stage can be toggled on or off from a single control panel group. <hr /><h2 id="why-two-passes">🎯 Why Two Passes?</h2>Anime-trained models respond far better to prompting than photo-trained ones. They produce a wider variety of images, follow tags more reliably, and give you greater creative control — but they can’t deliver true realism.Photo-trained models excel at fidelity and realism, but they’re harder to steer with prompts and tend to be narrower in range.⚡ The two-pass workflow bridges this gap:<ul><li>Pass 1 → Use an anime-style model for creativity, variety, and promptability.</li><li>Pass 2 → Re-render the Pass 1 result through a photorealistic model.<ul><li>IPAdapter + low denoise preserve identity and composition.</li><li>Face Detailer, upscalers, and swaps bring out realism.</li></ul></li></ul>✨ Extra flexibility:<ul><li>If you start with a good realism model, you can run only Pass 1.<ul><li>Adds, FreSca, and face tools still improve stability, lighting, and detail.</li><li>Pass 2 is optional — skip it when your realism model is already strong.</li></ul></li></ul><hr /><h1 id="workflow-control-panel-overview">⚡ Workflow Control Panel Overview</h1>The workflow is driven by a single control panel group that centralizes all the key decisions. The goal is to make it easy to explore ideas, iterate on renders, and then scale into multi-pass refinement without having to rewire or hunt for nodes.The panel does more than just toggle groups. It also provides full control over models, samplers, LoRAs, prompts, reference images, and post-processing.<hr /><h2 id="model-and-sampler-controls">🎛️ Model and Sampler Controls</h2>For both Pass 1 and Pass 2, you can set:<ul><li>Model checkpoint (separately for each pass)</li><li>Sampler (e.g. DPM++, Euler, etc.)</li><li>Scheduler (Karras, Simple, etc.)</li><li>Steps (iteration count per render)</li><li>CFG (guidance strength)</li><li>Denoise strength (how closely the pass follows the conditioning)</li></ul><hr /><h2 id="adds-strength-control">➕ Adds Strength Control</h2>The Adds section runs additional low-denoise passes on the same model to refine structure and variation.<ul><li>At strength = 1, it runs two passes at 0.16 and 0.8 denoise.</li><li>Increasing the number scales both denoise values proportionally.</li><li>This gives you fine-tuned control over how much extra shaping and diversity you inject after the base Pass 1 render.</li></ul><hr /><h2 id="reference-image">🖼️ Reference Image</h2><ul><li>A single Reference Image selector is in the control panel.</li><li>Required for ControlNet and RemBG features to activate.</li><li>Works hand-in-hand with the InPaint Fix toggle.</li></ul><hr /><h2 id="lora-integration">🎨 LoRA Integration</h2><ul><li>Uses EasyLoraStack for simple LoRA management.<ul><li>Defaults to three slots, but can be expanded.</li></ul></li><li>There’s also an unconnected Easy Load LoRA node.<ul><li>Right-click it → View LoRA Info → shows all the trigger tags that LoRA recognizes.</li></ul></li></ul><hr /><h2 id="prompt-and-seed-management">✍️ Prompt and Seed Management</h2><ul><li>Positive Prompt lives directly in the control panel.</li><li>Seed is set here and piped across the workflow for consistency.<ul><li>Designed to be fixed by default.</li><li>The Seed node has a “Manual Random Seed” button if you want variety.</li><li>With a fixed seed, you can incrementally add passes — iterate on Pass 1 until you like the base image, then activate Pass 2 and refiners for enhancement.</li></ul></li></ul><hr /><h2 id="face-related-controls">👤 Face-Related Controls</h2><ul><li>Face model selector for choosing your identity source model.</li><li>Face image input if you want to drive swaps from a real reference.</li><li>Toggle to decide whether Face Detailer should work on the raw Pass 2 image, or on the Pass 2 Face Swapped image (for tighter corrections).</li></ul><hr /><h2 id="gallery-extras">🖼️ Gallery Extras</h2>The Gallery is designed for quick inspection and control:<ul><li>Seed → clearly displayed for reproducing or iterating.</li><li>Face Index → lets you select which detected face is used for swaps.<ul><li><code>0 = largest face</code></li><li>Higher numbers = progressively smaller faces in the frame.</li></ul></li></ul><hr /><h2 id="fresca-(planned-addition)">🌀 FreSca (Planned Addition)</h2><ul><li>A control for FreSca frequency scaling will be added soon.</li><li>This will let you fine-tune global structure and texture detail separately, giving another layer of style control without rewriting prompts.</li></ul><hr /><h2 id="post-processing-(hotkey-4)">🎭 Post Processing (Hotkey 4)</h2>At the very end of the workflow, a Post Processing block applies finishing touches to whichever pass is active (Pass 1 or Pass 2):<ul><li>Expression Editor → tweak facial expressions.</li><li>Caption / Label → extend the image with a black bar containing custom text. Useful for annotation, dataset prep, or just keeping track of iterations.</li></ul><hr /><h1 id="complete-workflow-(everything-on)">📋 Complete Workflow (everything ON)</h1><ol><li>Reference Image input<ul><li>Feeds ControlNet and RemBG (with InPaint Fix).</li></ul></li><li>ControlNet conditioning (from reference).</li><li>RemBG processing (background removal / inpainting).</li><li>Pass 1 – Base render (model render with chosen sampler/scheduler/steps/CFG/denoise).</li><li>Pass 1 Adds (two low-denoise refinements).</li><li>Pass 1 Face Shape (InsightFace / Face Shaper).</li><li>Pass 1 Face Swap (Face Image or Face Model).</li><li>Pass 1 Upscale/Downscale refining.</li><li>Pass 1 Face Swap + Upscale/Downscale.</li><li>Pass 2 – Model render (IPAdapter FaceID, lower denoise).</li><li>Pass 2 Face Swap.</li><li>Pass 2 Face Detailer.</li><li>Pass 2 Upscale/Downscale refining.</li><li>Pass 2 Face Swap + Upscale/Downscale.</li><li>Pass 2 Face Detailer + Upscale/Downscale.</li><li>Expression Editor (facial tweaks, applied at the end).</li><li>Caption / Label (optional text overlay bar).</li><li>Pass 2 Gallery (final outputs, with Seed and Face Index).</li></ol><hr /><h1 id="quick-start">🏁 Quick Start</h1><ol><li>Load the workflow and check the Control Panel (Hotkey 1).</li><li>Select your Pass 1 model and set sampler, scheduler, steps, CFG, and denoise.</li><li>Enter your Positive Prompt (Hotkey 3) and fix a seed.</li><li>Run Pass 1 with or without Adds → this gives you a clean base render.</li><li>Iterate:<ul><li>If you like the base → toggle ON Pass 2 for advanced refinements.</li><li>Add Face Swap, Detailer, Upscale, Expression Editor, or Caption/Label as needed.</li><li>Check outputs in the Gallery (Hotkey 2) — Seed and Face Index are right there.</li></ul></li></ol><hr /><h1 id="dependencies">📦 Dependencies</h1>This workflow uses 12 well-supported node packs, most with 1500+ stars on GitHub and actively maintained (as of Sept 2025):<ul><li>ComfyUI_IPAdapter_plus</li><li>ComfyUI Impact Pack</li><li>ComfyUI Impact Subpack</li><li>ComfyUI-Custom-Scripts</li><li>ComfyUI-AdvancedLivePortrait</li><li>rgthree-comfy</li><li>ComfyUI-Easy-Use</li><li>ComfyUI-KJNodes</li><li>ComfyUI_FaceShaper</li><li>WAS Node Suite (Revised)</li><li>ComfyUI-RvTools</li><li>ComfyUI’s ControlNet Auxiliary Preprocessors</li></ul>✅ All dependencies are mainstream and up-to-date — no legacy or obscure node packs required.

DreamBridge 🌉 — bridging anime creativity to photo realism

# 🚀 Summary of New Functionality Since v10.4v15.4 represents a major evolution of the workflow — expanding from a static two-pass image pipeline into a modular image + video generation suite with enhanced realism, identity control, and automation.---## 🧩 Core Additions- Five Pass 1 render modes: Basic · Adds · 2-Stage · 3-Stage · Tensor-RT (fast mode) - Pass 2 engine toggle: choose between Flux or SDXL for different output aesthetics - SAM2 segmentation: replaces RemBG — removes or isolates any object, not just backgrounds - Face ID system: supersedes Face Shaper with multi-photo identity consistency - WANVideo 3-Stage pipeline: 1️⃣ Key-Frame taken from image pipeline 2️⃣ RIFE interpolation 3️⃣ Upscale + Color-Match finishing - Long-Video mode: stitches multiple clips into extended sequences - Tensor-RT acceleration: optional high-speed inference path for compatible GPUs ---## ⚙️ Pipeline Enhancements- Unified LoRA / ControlNet routing with on-the-fly bypass switches - FreSca detail pass now standardized across both image and video stages - Improved Gallery layout showing every major stage (image + video) - Fully re-labeled nodes and hotkey navigation (1–6) for faster workflow access

Files with SHA256: 52b8a5c7c054086e3ec9119958a579cc6f280411a56de9c7719eeee6337aebd2