Create realistic face swaps

Which models are the fastest?

Codeplugtech/face-swap model is one the fastest face swap models. It’s lightweight, runs on CPU, and often finishes predictions in under a minute. The easel/advanced-face-swap model is optimized for quality and multi-face workflows, so it may run a bit slower depending on resolution. The fofr/face-swap-with-ideogram model varies in speed depending on how much style or ideogram-driven processing is involved.

Which models offer the best balance of cost and quality?

For strong performance at a lower cost, the codeplugtech/face-swap model is a reliable choice and has been used a huge number of times on Replicate.
If you need more polished, higher-fidelity results, especially for professional or commercial use, the easel/advanced-face-swap model tends to be the most robust option.
Use fofr/face-swap-with-ideogram when you want stylistic or character-driven results rather than pure realism.

What works best for replacing one person’s face in a portrait with another?

For crisp, realistic portrait swaps, the easel/advanced-face-swap model usually produces the most natural results. It preserves lighting, skin texture, and pose especially well.
If you want something straightforward and quick, codeplugtech/face-swap handles simple portrait swaps with very little setup.
If you're swapping in a stylized or ideogram-derived identity, try fofr/face-swap-with-ideogram.

How about swapping faces for multiple people in the same image?

The easel/advanced-face-swap model supports swapping one or two faces at a time and is built to handle multi-person scenes reliably.
codeplugtech/face-swap can work on group photos too, but it may require more manual control to pick which face to swap.
For stylized or non-photorealistic group swaps, fofr/face-swap-with-ideogram is flexible but may require more experimentation.

What’s the difference between the approaches in this collection?

• Photorealistic swapping models: These take a source face and blend it into a target image while keeping lighting, expression, and background intact.
  Examples: easel/advanced-face-swap, codeplugtech/face-swap.
• Stylized or ideogram-based models: These let you perform identity swaps that also incorporate style, character, or brand-like consistency.
  Example: fofr/face-swap-with-ideogram.

Other differences include: number of faces supported, prompt control, resolution, and how well identity consistency is preserved across multiple images.

What kinds of outputs can I expect?

All models will return a target image with the source face integrated into the scene.
The easel/advanced-face-swap model outputs highly polished, commercial-grade swaps.
The codeplugtech/face-swap model outputs clean, reliable swaps that are great for prototyping or social uses.
The fofr/face-swap-with-ideogram model can generate more stylized or character-driven results.

How can I self-host or push a model to Replicate?

To publish your own face-swap model, you containerize your inference code, define an input schema (e.g., source_image, target_image, options), and upload it to Replicate.
Once it’s published, you can run it just like the models in this collection and share it publicly.

Can I use these models for commercial work?

Yes, but always check the License section on each model’s page to confirm. Also remember that likeness rights and consent matter whenever you’re swapping real faces.

How do I run these models?

Upload:

• A source face (the face you’re bringing in)
• A target image (where the face goes)

Then set any parameters (face index, output size, etc.) and run the model.
For example:

• codeplugtech/face-swap is great for quick swaps.
• easel/advanced-face-swap gives you more control and detail.
• fofr/face-swap-with-ideogram allows stylized identity control.

What should I know before running a job in this collection?

• Clear, well-lit source faces work best.
• Try to match the angle of the face in the target image.
• Higher resolution costs more and takes longer.
• Some models allow you to specify which face to swap in multi-person scenes.
• Test a few images to dial in your workflow before running lots of predictions.

Any other collection-specific tips?

• If you want realistic, commercial-ready swaps, use easel/advanced-face-swap.
• If you want fast, efficient swaps for social content, use codeplugtech/face-swap.
• If you're building stylized characters or consistent identities, try fofr/face-swap-with-ideogram.
• Always consider privacy and consent when swapping real faces.

What if I want to automate face-swap workflows in an app?

All three models work well in automated pipelines.
For high-volume apps, pick a model with predictable runtime like codeplugtech/face-swap.
For premium image-editing or photo-shoot workflows, integrate easel/advanced-face-swap.
If your app involves stylized avatars or character identity systems, use fofr/face-swap-with-ideogram.