01
Choose the source face
A sharp, unobstructed face works best. Sunglasses and heavy tilt reduce quality.
Démo publique gratuite
FaceFusion Face Swap
Envoyez un visage source et une photo cible. FaceFusion change l’identité tout en conservant pose, lumière et expression.
Gratuit pour les visiteurs Pas de clé API Démo CPU · images
Outil en direct
Échanger un visage
Visage source + image cible en entrée, PNG en sortie. Le studio simple sert aux essais rapides. L’onglet FaceFusion complet expose HyperSwap, INSwapper et les améliorations.
Loading the full FaceFusion demo. Sleeping apps can take a minute to wake.
What it is
FaceFusion is a face-swap pipeline, not a video generator
FaceFusion takes a source identity and maps it onto a target image. HyperSwap and INSwapper are the swap models; face enhancers clean skin and edges after the swap.
Wan-Animate and Wan 2.2 S2V generate motion from a still. FaceFusion keeps the target photo and only changes who is in it. Use this page for identity transfer, not talking or singing videos.
How to swap a face
Three steps: a source identity, a target photo, then optional enhance.
02
Choose the target photo
The output keeps this pose, crop, and lighting. One clear face is more reliable than a crowd.
03
Swap, then enhance
The studio runs a CPU face swap. Turn on enhance to sharpen the result. The full tab exposes more models.
Models in this demo
The simple studio uses a streamlined FaceFusion path. The full tab can reach HyperSwap and INSwapper.
HyperSwap 1A / 1B
FaceFusion’s newer swap family. Strong identity transfer on stills; licensed ResearchRAIL.
INSwapper
A widely used InsightFace swapper. Fast and familiar, with non-commercial limits on some weights.
Face enhancer
Optional cleanup after the swap so skin, teeth, and edges look less pasted.
What people actually swap
The same source + target flow covers portraits, characters, restorations, and light creative edits.
Portrait identity
Put one face onto another headshot while keeping camera angle and wardrobe.
Character stills
Try a face on illustrated or styled characters when the face region is large enough.
Photo experiments
Preview how a person would look in an existing photo before a real shoot.
Playful edits
Quick joke swaps. Keep it consensual — do not upload other people’s private photos.
FaceFusion vs Wan-Animate
Both start from a face, but they solve different jobs.
| Feature | FaceFusion | Wan-Animate / S2V |
|---|---|---|
| Job | Replace identity on an existing still | Generate a new talking or moving clip |
| Inputs | Source face + target image | Image + motion video or audio |
| Motion | None in this image demo | Full performance and lip motion |
| Best for | Face swap, identity transfer | Talking photos, character animation |
Better swap tips
The model copies identity. Matching pose and lighting does more than any extra slider.
- 01. Use a front face for the source. Profile-to-profile is harder than frontal-to-frontal.
- 02. Keep similar head size and crop between source and target.
- 03. Avoid tiny faces, heavy occlusion, and group shots for the first try.
- 04. If the result looks plastic, retry with enhance on, or a closer source crop.
FaceFusion FAQ
Is this FaceFusion demo free?
Yes for visitors. It uses public CPU FaceFusion backends, so there is no GPU bill on this page. Queues and downtime still happen.
Is this unlimited ZeroGPU?
No. Ordinary ZeroGPU Spaces have short daily GPU minutes. This demo prefers a CPU FaceFusion Space that does not spend ZeroGPU quota, so it is slower but better for a small public try-out.
Can I swap video here?
The simple studio is image-only. The full FaceFusion tab may expose more processors, but this page is built and tested as a still-image demo.
Why is it slow?
CPU inference plus a public queue. A 480p-ish crop with one face is kinder than a huge photo.
Can I sell a Face Swap product on this backend?
No. Public Spaces have no SLA, and HyperSwap / INSwapper licenses are not a commercial SaaS grant. Production later means a licensed API or your own GPU.
What photos should I avoid?
Anything private, anyone who did not consent, and any illegal or harmful imagery. Uploads go to a third-party public demo.