FaceFusion gives you multiple face swap models to choose from, and Inswapper and HyperSwap are the ones people ask about most. The core difference is resolution: Inswapper runs at 128x128 pixels, while HyperSwap runs at 256x256 pixels. That's 4x more pixels, which means noticeably sharper facial details before you even apply any enhancement.
HyperSwap also comes in multiple variants optimized for different scenarios. But higher resolution means more VRAM and longer processing time. Inswapper is still a solid choice when you need speed or when you're working with limited GPU resources.
What you might be experiencing
- "Inswapper looks blurry even after enhancement."
- "Which HyperSwap model should I use—1A, 1B, or 1C?"
- "HyperSwap is sharper but takes forever to process."
- "Why does my swap look plasticky when I zoom in?"
- "Is HyperSwap worth the extra VRAM?"
- "Inswapper works but the face looks painted."
- "What's the difference between 128px and 256px models?"
If you're wondering about any of these, let's break it down.
When model choice matters most
Choosing the right model matters most in these scenarios:
-
High-resolution output: When your final video is 1080p or 4K, Inswapper's 128px face patch gets visibly soft after upscaling. HyperSwap's 256px output holds up much better at larger frame sizes.
-
Close-up shots: When faces take up a big chunk of the frame, resolution limits show up fast. HyperSwap gives you finer detail in eyelashes, skin texture, and lip definition.
-
Reducing enhancer dependency: If you want natural-looking results without cranking up the enhancement, HyperSwap's higher native resolution helps. Less hallucinated detail means fewer plastic artifacts.
-
Limited VRAM: HyperSwap needs more GPU memory than Inswapper. On GPUs with 4-6GB VRAM, Inswapper might be your only practical option.
-
Batch processing large videos: When you're grinding through hours of footage, Inswapper's speed advantage might outweigh HyperSwap's quality advantage.
Why this matters
1. Resolution is the main difference
| Model | Resolution | Relative Detail |
|---|---|---|
| inswapper_128 | 128x128 | Baseline |
| HyperSwap 1A/1B/1C | 256x256 | 4x more pixels |
More pixels = more facial information preserved and transferred. When a 128px face gets upscaled to fit a 1080p frame, it's being enlarged roughly 15x. At 256px, that enlargement drops to about 7.5x—a big reduction in interpolation artifacts.
2. HyperSwap has three variants
FaceFusion introduced three HyperSwap variants in version 3.3.0, each tuned for different situations:
| Variant | Best for |
|---|---|
| HyperSwap 1A | Handles occlusion and tricky angles well; good with makeup |
| HyperSwap 1B | Balanced general-purpose model for most use cases |
| HyperSwap 1C | Maintains head shape accurately; behaves more like traditional Inswapper |
Which one should you pick? Start with 1B—it's the safest default. Use 1A if you're dealing with partially hidden faces. Use 1C when preserving the original head proportions is critical.
3. The enhancer compensation problem
Because Inswapper's output is inherently low-resolution, people often crank up face enhancement to compensate. This creates that familiar plastic or over-smoothed look. HyperSwap reduces this dependency—the higher base resolution means you can use lighter enhancement or skip it entirely, preserving more natural texture.
4. Performance and VRAM costs
Higher resolution models need more GPU memory and take longer to process:
- Inswapper is faster and uses less VRAM
- HyperSwap needs more VRAM and takes longer per frame
For real-time or near-real-time work, Inswapper might be necessary. For offline processing where quality matters more than speed, HyperSwap is usually the better choice.
5. Why Inswapper still exists
Inswapper isn't obsolete. It's widely supported across multiple tools—Roop, Reactor, and other face swap systems all use it. Its low resource requirements make it accessible on older or weaker hardware. For many use cases—low-res output, distant faces, quick prototyping—Inswapper is still perfectly fine.
Trade-offs you'll face
-
Quality vs speed: HyperSwap produces better results but processes more slowly than Inswapper.
-
VRAM vs resolution: HyperSwap needs more GPU memory. If you're short on VRAM, you might be stuck with Inswapper.
-
Native detail vs enhancement: HyperSwap reduces your reliance on face enhancers, potentially giving more natural results. Inswapper usually needs enhancement to compensate for its low resolution.
-
Model variants vs simplicity: HyperSwap's three variants give you flexibility but require understanding which to use. Inswapper is just one model—no choices to make.
-
Compatibility vs features: Inswapper works across many tools. HyperSwap is specific to FaceFusion 3.3.0+.
Frequently asked questions
Q: Which model should I use by default? A: If you have enough VRAM and quality matters, start with HyperSwap 1B. If you need fast processing or have limited GPU resources, Inswapper works fine.
Q: What's the difference between HyperSwap 1A, 1B, and 1C? A: 1A handles occlusion and angles better. 1B is balanced for general use. 1C preserves head shape more accurately. Start with 1B and switch if you run into specific issues.
Q: Can I use HyperSwap without a face enhancer? A: Yes. HyperSwap's 256px output is often sharp enough to skip enhancement entirely, especially for 1080p output or when faces are smaller in the frame.
Q: Why does Inswapper look painted or plasticky? A: The 128px resolution just doesn't capture fine detail. When upscaled and enhanced, the result often looks artificial because the enhancer is making up detail that was never there.
Q: Is HyperSwap available in other tools like Roop? A: No. HyperSwap models are developed by FaceFusion Labs and currently only available in FaceFusion. Other tools typically use Inswapper or similar 128px models.
Related articles
- Understanding FaceFusion Face Enhancers – How enhancement interacts with model resolution
- Why AI Face Swaps Can Look Blurry or Soft – The underlying resolution bottleneck
- Why FaceFusion Settings Feel Ineffective – Why model selection matters more than slider adjustments
Final thoughts
The choice between HyperSwap and Inswapper mostly comes down to resolution. HyperSwap's 256px models produce visibly sharper results and reduce your dependence on enhancement, but they cost more in resources. Inswapper remains practical for quick work, limited hardware, or when faces are small in the frame. Understanding this trade-off helps you pick the right tool instead of defaulting to one model and wondering why your results vary.