Model comparison

The cheat sheet

One glance at the trade-offs. VRAM is qualitative (hardware and version dependent); treat it as a relative ordering, not a number.

Swappers

The model that replaces the face region.

Model	Res	VRAM	Pros	Cons	Best for
inswapper_128	128	mid	+ Best out-of-the-box identity likeness + Robust across angles/lighting	- Soft at high zoom - 128px latent limits fine detail	Default for most swaps; pair with an enhancer for crisp close-ups.
simswap_256	256	low	+ Good identity from one clean source + Lower VRAM than larger nets	- More sensitive to occlusion/pose - Smaller ecosystem tuning	Single clean source image, modest hardware.
simswap_512recent	512	mid	+ Sharper detail than 256 + Good embedding-driven likeness	- Heavier on VRAM/throughput	Embedding-driven swap where you want more resolution.
ghost_256	256	low	+ Tolerates occlusion/cropped faces + Fast throughput	- Less identity fidelity than inswapper in clean conditions	Fast work, partially occluded or moving subjects.
uniface_256recent	256	mid	+ Newer, actively maintained + Good identity/speed balance	- Not present in older releases	Newer installs wanting a modern default swap.

Enhancers

Restore and sharpen the enhanced face.

Model	VRAM	Pros	Cons	Best for
codeformer	mid	+ Tunable fidelity via weight + Strong general restoration	- Can plasticize skin at high weight - Needs tuning per face	Best all-round face enhancer; the default first choice.
gfpgan_1.4	mid	+ Punchy detail/sharpening + Well documented	- Can over-smooth or hallucinate at low weight	Trying for bold sharpening; gfpgan_1.2 where 1.4 is unavailable.
restoreformerrecent	mid	+ Clean, few artifacts + Good for degraded/noisy input	- Slightly softer than gfpgan at extremes	Noisy or low-quality source frames.
restoreformer_plus_plusrecent	mid	+ Better detail than base restoreformer	- Heavier checkpoint	Preferred when restoreformer is your enhancer.
gpen_bfr_512	mid	+ Very explicit detail at high res + Strong on heavily degraded input	- High VRAM at 1024/2048 - Can over-texture	High-res pipelines with VRAM to spare.
gpen_bfr_1024	high	+ Fine detail for close-ups	- High VRAM	4K+ close-up work.
gpen_bfr_2048	high	+ Extreme detail	- Very high VRAM - Slow	High-VRAM only, large output resolution.

Detectors

Find faces before anything else runs.

Model	VRAM	Pros	Cons	Best for
retinaface	mid	+ Accurate, good recall + Handles varied scales	- Heavier than yoloface	Balanced default detection.
SCRFDrecent	low	+ Fast + accurate + Good on mobile/laptop	- Newer, less historical default	Speed + accuracy on real hardware.
yoloface	low	+ Very fast, low memory	- Lower recall on small/occluded faces	Low-VRAM and fast scans.

Quick picks

Not sure where to start? inswapper_128 + codeformer covers most general swaps. Choose ghost_256 when occlusion or fast motion hurts the swap, yoloface when VRAM is tight, and jump to gpen_bfr_1024+ or a second frame_enhancer pass only when you have GPU headroom and need the detail.