Skip to content
Other

GRACE Deepfake Detection

Explainable Deepfake Detection System

2 min readupdated

The GRACE (Generalisable, Robust And Concept-Explainable detection) model addresses the task of detecting fake faces in videos and images by scoring them across ten clear forensic features. Instead of simply outputting a "real or fake" verdict, the system provides a detailed breakdown by concepts (such as skin artifacts, color mismatches, or geometry inconsistencies). The architecture includes a heavy teacher model based on CLIP ViT-L/14 with LoRA adapters and a compact EfficientNet-B0 student for fast CPU inference. Created as part of an academic capstone project at PES University, it is useful for security researchers and developers building media verification systems.

Key features

  • Ten concepts: Evaluation based on specific features like blur, blending artifacts, or eye editing.
  • Teacher-student architecture: An accurate teacher for powerful servers and a lightweight 4.7-million-parameter ONNX model for CPUs.
  • Robust base: Trained on the FaceForensics++ dataset with self-blended fakes without using test data.
  • Ready-to-use interface: Source code and a demo space available on Hugging Face Spaces.

What you can do

  • Media verification: Run the built-in demo to analyze suspicious videos and photos for manipulations.
  • CPU integration: Use the lightweight grace_student.onnx version for fast video analysis on standard servers.
  • Vulnerability analysis: Examine model weights and configurations to research the reliability of modern deepfake detectors.

Example

The model evaluates 224×224 pixel faces across ten forensic criteria, outputting an interpretable fake logit as the weighted sum of these concepts.

Read moreDetails
AUC metrics: FF++ 0.997, DFD 0.951, Celeb-DF 0.901
Limitations: Vulnerable to new generators (diffusion models), heavy video compression, and extreme head pose angles
Components: PyTorch teacher weights with LoRA on CLIP and a compact EfficientNet-B0 student
Discuss Telegram 0

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 0.0 GB · ready files0.5 GB
  • FP8 / INT8recommendedweights 0.0 GB0.5 GB
  • 4-bitweights 0.0 GB0.5 GB

Tip: go 8-bit.FP8 / INT8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: FP8 / INT8 — from RTX 2060 · BF16 — from RTX 2060

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # grace-cap/grace-deepfake-models
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download grace-cap/grace-deepfake-models --local-dir ~/models/grace-deepfake-models
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to grace-deepfake-models
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

New posts about versions of grace-deepfake-models (quants, finetunes, LoRAs) will show up under the bell.

No known relatives of this model yet. All derivatives on Hugging Face

About the model

grace-cap/grace-deepfake-models
Author
grace-cap
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

00

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments