Face Swap - SX CU

🎭 Cinematic Face Swap – ComfyUI Workflow

Created by Me – a filmmaker & AI workflow designer. This is my custom tool for realistic, identity‑preserving face swapping in cinematic scenes, and I’m sharing it on Civitai for creators who need precise, controllable facial replacement.

🧩 Overview

This advanced ComfyUI workflow takes a base image (the target scene) and one or two reference face images, then seamlessly replaces the face(s) in the base image while preserving lighting, expression, head angle, micro‑expressions, and any facial details (like dirt, water, or scars). It supports both single‑face and couple‑face swaps with a single toggle.

1️⃣ Required Models & Prerequisites

Make sure these files are placed in your ComfyUI models/ folders:

FolderRequired Filesdiffusion_models/qwen_image_edit_2511_bf16.safetensors (or fp8 version)loras/Qwen-Image-Edit-2511-Lightning-4steps-V1.0-bf16.safetensors
Qwen_bfs_head_v5_2511_merged_version_rank_16_fp16.safetensors (face‑swap LoRA)text_encoders/qwen_2.5_vl_7b_fp8_scaled.safetensors
gemma4_e4b_it_fp8_scaled.safetensors (optional, for auto‑prompting)vae/qwen_image_vae.safetensors

⚠️ VRAM Requirement: ~12–16 GB recommended because the workflow loads Qwen, Gemma, and the face‑swap LoRA simultaneously.

3️⃣ Automatic Prompt Generation

The workflow uses Gemma 4 (optional) to analyse the input images and build a detailed prompt that instructs the model exactly what to preserve and what to replace:

• Single‑face mode: Replaces the head in Image 1 with the head from Image 2, while strictly keeping identity features (hair, eyes, nose, lips) from Image 2, but expression, head angle, and micro‑expressions from Image 1.

• Couple‑face mode: Replaces two different heads (e.g., a woman and a man) in Image 1 with faces from Image 2 (woman) and Image 3 (man), again preserving identity from the source images and expression from the base image.

You can also write your own custom prompt – the workflow gives you full control via a simple string input.

6️⃣ Advanced Settings (Optional)

• Same Size Toggle: Enable this to automatically resize the output to match the base image’s resolution – or set a custom width manually.

• Steps & CFG: Defaults are 4 steps and 1.0 CFG (Lightning‑optimised). You can increase them for more detail.

• Couple Switch: Flip this to instantly change between single‑face and two‑face swap without changing anything else.

• Additional Prompt: Append extra instructions (e.g., “keep the background sharp” or “make the skin tone match perfectly”).

📦 Final Output

After running, the workflow saves the result as:

• Qw_Face_*.png (or your custom prefix) – a high‑quality, seamless face‑swapped image with the original lighting and composition fully preserved.

Why I built this: In filmmaking, I often need to test different actors or versions of a character without reshooting. This workflow lets me swap faces in seconds while maintaining the emotional context of the original performance – perfect for pre‑visualisation, casting tests, or even final shots. I hope it saves you time and unlocks new creative possibilities! 🎥✨