System Requirements

Hardware and software prerequisites for running Recaster.

Recaster runs on macOS, Windows, and Linux. The hardware requirements depend on whether you plan to process locally on your own GPU or use the Studio tier to offload work to cloud GPUs. This page covers both scenarios.

Minimum Requirements

These are the bare minimum specs needed to run Recaster. Performance may be limited, especially for video processing and training.

Component	Minimum
Operating System	macOS 12 (Monterey)+, Windows 10 (64-bit), Ubuntu 20.04+
CPU	64-bit processor, 4 cores
RAM	8 GB
Storage	5 GB free space
GPU	NVIDIA GPU with 4 GB VRAM and CUDA support
Display	1280 × 720 resolution

Recommended Specs

For the best experience with video processing, model training, and AI upscaling, we recommend the following hardware:

Component	Recommended
CPU	Modern 8-core processor (Intel i7/i9, AMD Ryzen 7/9, Apple M1 Pro+)
RAM	32 GB
Storage	SSD with 50 GB+ free space
GPU	NVIDIA RTX 3060 or better with 8 GB+ VRAM
Display	1920 × 1080 or higher

SSD Strongly Recommended

Face extraction, training, and video processing involve reading and writing thousands of image files. An SSD makes a significant difference in overall performance compared to a traditional hard drive.

GPU Compatibility

GPU acceleration is essential for practical face processing speeds. Recaster uses different GPU frameworks depending on the platform and workload.

NVIDIA GPUs (CUDA)

NVIDIA GPUs with CUDA support are the primary acceleration target. Most features use ONNX Runtime with the CUDA execution provider. Training uses TensorFlow with CUDA.

Supported: GTX 10-series and newer (GTX 1060+, RTX 20/30/40 series)
Best performance: RTX 3060, RTX 3080, RTX 4070, RTX 4090
VRAM recommendation: 8 GB for Quick Recast, 12 GB+ for training

Apple Silicon (macOS)

On macOS with Apple Silicon (M1, M2, M3, M4), Recaster uses CoreML acceleration for some inference models like Real-ESRGAN. However, training models locally requires an NVIDIA GPU, so macOS users who want to train should consider the Studio tier for cloud GPU access.

CoreML Limitation

SwinIR upscaling models are not compatible with CoreML due to dynamic input shapes. On macOS, SwinIR automatically falls back to CPU processing. For best upscaling performance on Mac, use Real-ESRGAN (CoreML accelerated) or remote upscaling via Studio tier.

AMD & Intel GPUs

AMD and Intel GPUs are not currently supported for GPU acceleration. Processing will fall back to CPU mode, which is significantly slower. If you have an AMD or Intel GPU, consider using the Studio tier for cloud processing on NVIDIA hardware.

RTX 5000 Series Compatibility

RTX 5000 Series GPUs

NVIDIA RTX 5090, 5080, 5070, and 5060 GPUs use the Blackwell architecture (SM 12.0). TensorFlow dropped native Windows GPU support after version 2.10, and current builds do not support compute capability 12.0. This means local training is not supported on RTX 5000 series GPUs on Windows.

If you have an RTX 5000 series GPU, Recaster will display a compatibility warning on startup. Here are your options:

Remote Training (recommended): Use the Studio tier to train on cloud GPUs via Vast.ai. This works immediately with no local GPU dependencies.
WSL2 Workaround: Use Windows Subsystem for Linux 2 with a community DeepFaceLab fork that supports newer architectures.
Quick Recast: Face swapping and enhancement via ONNX Runtime may work on RTX 5000 series, as ONNX Runtime receives faster updates for new GPU architectures.

Dismiss the Warning

If you are aware of the RTX 5000 limitation, check "Don't show again" on the startup warning dialog. This setting is saved in your preferences and persists across launches.

Software Dependencies

Recaster bundles most of its dependencies in the installer. The following external software is optional but recommended:

Software	Required?	Purpose
FFmpeg	Recommended	Audio preservation in video processing, video format conversion
NVIDIA Drivers	Required for GPU	CUDA support for GPU-accelerated processing
rsync	Studio tier	File synchronization with remote instances. Pre-installed on macOS and Linux. Available via Git Bash on Windows.

Cloud GPU Alternative

Studio Tier Cloud GPUs

With the Studio tier, you can offload training, Quick Recast processing, and video upscaling to cloud GPUs via Vast.ai. This means your local hardware requirements are much lower — you only need enough power to run the Recaster interface itself. Cloud instances typically use RTX 3090 or RTX 4090 GPUs with 24 GB VRAM.

If your local hardware does not meet the GPU requirements, the Studio tier is an excellent option. You can run the Recaster interface on any machine (even a laptop without a dedicated GPU) and process on powerful cloud GPUs. See the Cloud Instances documentation for more details.

Was this page helpful?