Roop Is Dead. ReActor Just Showed Up.
The NEW KING of Face Swappers! Reactor Tutorial. Roop is dead.
Tools break. Projects get abandoned. The thing you built your whole workflow around vanishes overnight. Sound familiar? If you've been doing AI face swapping with Roop in Stable Diffusion, you already know this feeling. But here's the good news... the replacement isn't just adequate. It's better.
Sebastian Kamph walks through the full setup and workflow for ReActor, the extension that stepped into the gap Roop left behind. And stepped up.
ReActor brings high-resolution face swaps with built-in upscaling. It runs on CPU only... no beefy GPU required. It supports both SDXL and SD 1.5 models. It auto-detects gender and age. And the part that matters most for longevity? It's still actively developed.
Roop gave us a lot. But Roop stopped moving forward. ReActor picked up the torch and kept running.
Getting It Running
Before anything creative happens, you need the foundation in place. ReActor requires Visual Studio 2022... specifically the C++ Build Tools. You can grab the full Community edition or just the lightweight build tools. Either works. The full version gives you more flexibility for future Stable Diffusion extensions, so if you're planning to tinker long-term, go big.
Once Visual Studio is installed, head into Automatic1111 WebUI. Navigate to Extensions > Install from URL, paste the ReActor GitHub repository link, and hit install. Restart the UI. If your terminal reads "ReActor status running"... you're in business.
Smooth process. A few minutes of patience. Then the real fun starts.
The Simplest Swap: img2img with Zero Denoising
The fastest way to see ReActor work is with an existing image. Drop your target photo into the img2img tab. Click the little ruler icon to match dimensions. Set denoising strength to zero... this is critical. You're telling Stable Diffusion to change nothing about the image itself. The only transformation comes from ReActor.
Open the ReActor panel. Upload your source face. Enable it. Generate.
That's it. BAM, new face on the same body. Sebastian demonstrates this by swapping a woman's face with the Mona Lisa. The result is clean, fast, and surprisingly seamless for something that took about four clicks.
Targeting Specific Faces in Group Photos
This is where it gets genuinely useful. Got an image with multiple people? ReActor numbers each detected face from left to right... 0, 1, 2, 3, and so on.
Want to swap only the second person? Set the target to 1. Want to swap two people simultaneously with different source faces? Map the numbers. Source face A goes to position 0, source face B goes to position 1. You can even flip them... put person A's face on person B's body and vice versa.
Sebastian demonstrates this with Johnny Depp's face on a two-person image. Change the target number, change which person gets the swap. Simple numbering system. Powerful control.
For anyone doing character work, composite scenes, or creative projects with multiple subjects... this feature alone justifies the switch from Roop.
Integrating Face Swaps into txt2img Generation
Here's where the workflow shifts from correction to creation.
Instead of generating an image and then fixing the face afterward, you can have ReActor apply your chosen face during the initial txt2img generation. Write your prompt... "man close-up portrait" with cinematic styling, for example. Load your source face into ReActor. Enable it. Generate.
Stable Diffusion creates the image. Another face appears first. Then ReActor steps in and swaps it with yours before the final output. You're not post-processing. You're building the face into the creative pipeline from the start.
Sebastian runs a batch of four images this way. Three male portraits and one female... all carrying his facial features. The system handles gender differences gracefully because of that automatic detection working in the background.
The Upscaling Order Matters
Small detail. Big impact.
ReActor includes built-in upscaling options. Sebastian recommends NMKD Siax or R-ESRGAN 4x+ as solid choices. The key insight: upscale after the face swap, not before. Reversing the order degrades quality.
He also specifically recommends enabling CodeFormer face restoration when using ReActor. In general Stable Diffusion workflows, face restoration can be hit-or-miss. But paired with ReActor's swap process, CodeFormer consistently improves results.
Tools working together. Each one doing what it does best. The right sequence creates the right outcome.
Why This Matters Beyond Face Swapping
Every tool eventually dies or gets abandoned. The ones that survive aren't always the originals... they're the ones that showed up when the originals couldn't anymore.
ReActor didn't just replace Roop. It added CPU support for people without expensive hardware. It added SDXL compatibility for people working with newer models. It added active development for people who need to trust their tools will keep working tomorrow.
That pattern matters whether you're talking about software extensions or anything else in life. The reliable replacement that keeps showing up beats the brilliant original that stopped.
Your workflow doesn't have to die when your tools do. ReActor is proof that the next chapter can be stronger than the last one. Install it. Experiment with it. Break something, learn something, build something better. And if you've been sitting on the sideline since Roop went dark... the light just showed up. ✨
--- Source: https://www.youtube.com/watch?v=Da6J3wjBx3A
From TIG's Notebook
Thoughts that surfaced while watching this.
Living the lives we want not only requires doing the right things but also necessitates not doing the things we know we'll regret. — *Nir Eyal, Indistractable*— TIG's Notebook — Core Principles
title: Quotes & Stats - TIG izms
That's the funny thing about hope. Nobody else gets to decide if you feel it. That choice belongs to you. — *K-Pop Demon Hunters*— TIG's Notebook — Core Principles
Echoes
Wisdom from across the constellation that resonates with this article.
A comedy maker builds three genuinely practical studio projects, revealing principles about finding shared patterns in chaos, tracking time as awareness instead of pressure, and using every available tool to close the gap between imagination and reality.
The asymmetry between the quality of Perplexity's execution and the fragility of its structural position, that is the thing I can't stop thinking about.
Three simultaneous shifts in consumer hardware, agent attention spans, and AI memory systems are converging in 2026 to make AI agents viable for mainstream adoption.