Quick Recast Basics

Drag-and-drop face replacement in minutes.

Quick Recast is Recaster's primary face replacement tool. It provides a streamlined interface where you drag and drop a source face and target media, select your preferred models, and let the processing pipeline handle everything else. This page covers the core interface and workflow.

The Interface

Open Quick Recast by clicking the Quick Recast icon in the left sidebar (lightning bolt icon). The interface is organized into several zones:

Left Side: Source

Source face drop zone — drag an image here
Face thumbnail preview after loading
Multiple source support for multi-identity mode

Right Side: Target

Target drop zone — drag image or video here
Target preview with detected face indicators
Video trim controls (when target is a video)

Bottom: Controls & Preview

Model selection dropdowns (swap model + enhancer)
Process button and progress bar
Before/after comparison slider for results
Output settings (file location, format)

Source & Target

Source Face

The source face is the identity you want to transfer onto the target. Drag an image file containing a clear, front-facing photo of the source person onto the Source drop zone.

Use a high-quality, well-lit photo with the face clearly visible
Front-facing or slight angle works best
The face should occupy a significant portion of the image
Avoid photos with heavy makeup, sunglasses, or face obstructions

Best Source Photos

For the highest quality results, use a source photo where the face resolution is at least 256x256 pixels. Photos from professional headshots, passport photos, or well-lit selfies produce the best results. Multiple angles of the same person are not needed for Quick Recast — a single clear photo is sufficient.

Target Media

The target is the image or video where you want to replace a face. Drag it onto the Target drop zone. Quick Recast supports:

Single images — JPEG, PNG, BMP, WebP
Videos — MP4, AVI, MOV, MKV (frame-by-frame processing)

When a target is loaded, Recaster automatically detects all faces in the first frame. If multiple faces are detected, you can choose which face to replace (or use multi-identity mode to replace multiple faces).

Swap Models

Quick Recast supports multiple face swap models with different characteristics. Select a model from the Swap Model dropdown in the settings panel. Models marked with a lock icon require a Studio license.

Name	Tier	Variants	Quality	Description
InSwapper 128	Free	128	Good	Standard ONNX Runtime face swap model. Reliable results for most use cases with good speed.
InSwapper 128-FP16	Free	128 (half-precision)	Good	Half-precision variant of InSwapper. Faster processing with slightly reduced precision.
Ghost	Studio	1, 2, 3	Very High	ArcFace-based face swap with high identity preservation. Three variants with increasing quality.
BlendSwap	Studio	256	Highest	Advanced blending technology at 256px resolution. Best for close-up shots and high-detail work.

InSwapper (Free Tier)

InSwapper is the default face swap model and is available on the Free tier. It uses ONNX Runtime for inference and produces good results for most scenarios. The 128 variant operates at 128x128 face resolution.

InSwapper 128: Standard precision. Best quality among Free tier options.
InSwapper 128-FP16: Half-precision (float16) variant. Approximately 20-30% faster with minimal quality difference. Good choice for previewing or processing long videos.

Ghost (Studio Tier) Studio

Ghost is a high-quality face swap model that uses ArcFace embeddings for superior identity preservation. It comes in three variants:

Ghost 1: Balanced quality and speed
Ghost 2: Improved detail preservation
Ghost 3: Highest quality, most compute-intensive

BlendSwap (Studio Tier) Studio

BlendSwap uses advanced blending technology at 256x256 resolution, twice the resolution of InSwapper. This model excels at close-up shots where facial detail is critical and produces the most natural-looking blends.

Automatic Model Download

All models are automatically downloaded from Hugging Face on first use. The download happens in the background with a progress indicator. After the initial download, models are cached locally and load instantly on subsequent uses.

Before/After Preview

After processing, Quick Recast displays a before/after comparison view. An interactive slider lets you drag left and right to reveal the original and processed versions side by side. This is useful for evaluating swap quality and checking for artifacts.

For video processing, the preview updates in real-time as frames are processed. You can see intermediate results without waiting for the entire video to finish.

Processing

Once your source and target are loaded and your model is selected, click the Process button to start face replacement.

Image Processing

For single images, processing is nearly instant (typically 1-3 seconds depending on your GPU). The result appears in the preview area immediately.

Video Processing

For videos, processing happens frame by frame. A progress bar shows the current frame number and percentage. You can stop processing at any time with the Stop button. For full details on video-specific features, see the Video Processing guide.

Output Settings

By default, Quick Recast saves the processed result alongside the original target file with a _recast suffix. For example, processing target.mp4 produces target_recast.mp4.

You can customize the output location by clicking the folder icon next to the output path field.

Detection Quality

The face detection step determines how accurately faces are found in the target image or video. You can adjust the detection quality from the settings panel:

Setting	Resolution	Speed	Best For
Fast	320px	2-3x faster	Close-up faces, previews, batch processing
Balanced	640px	Standard	General use, good accuracy and speed tradeoff
Accurate	1024px	Slower	Small/distant faces, group shots, maximum accuracy

Default is Fast

Detection quality defaults to Fast (320px) for the best processing speed. If you notice that small or distant faces are being missed, switch to Balanced (640px) or Accurate (1024px). This setting is especially important for group scenes with many faces at different distances.

Remote Processing Studio

Studio tier users can process on cloud GPUs for significantly faster results, especially with large videos. Toggle the Remote switch in the Quick Recast interface to enable cloud processing.

When remote mode is enabled, Quick Recast uploads your source and target to a cloud GPU instance, processes the video there, and streams preview snapshots back in real time. The final result is downloaded when processing completes.

For details on setting up cloud instances, see the Cloud Instances documentation.

Content Safety

Quick Recast includes a built-in content safety system that checks both source and target faces before processing. The system uses a Challenge-25 policy: if any detected face appears to be under 25 years old, processing is blocked. This is a non-configurable safety measure designed to protect minors.

If processing is blocked, a dialog will explain the reason. This check uses the FairFace classifier for age estimation, which works with age ranges rather than exact ages. The system errs on the side of caution.

Was this page helpful?