belkhir-nacim/deepfake-videos

Unified Deepfake Video Dataset

913,830 videos from 22 datasets, unified for deepfake detection research.

📦 Contains metadata + actual video files under CC-BY-4.0, with permission from all original dataset authors.

Metric	Value
Total videos	913,830
Real videos	339,405
Fake videos	574,425
Generated	2026-04-22T17:09:30Z

Quick Start

from datasets import load_dataset

ds     = load_dataset("belkhir-nacim/deepfake-videos")
sample = ds["train"][0]
print(sample["label"])          # "real" or "fake"
print(sample["dataset_source"]) # e.g. "UADFV"
print(sample["hf_path"])        # "videos/UADFV/fake/0001.mp4"

# Filter
fakes  = ds["train"].filter(lambda x: x["label"] == "fake")

# Stream (no full download)
stream = load_dataset("belkhir-nacim/deepfake-videos", streaming=True, split="train")
for sample in stream:
    process(sample)

Dataset Coverage

Dataset	Total	Real	Fake	Status
DDL_dataset	403,702	196,112	207,590	🔗 HF source
DFD	3,431	363	3,068	✅ HF-hosted
DeepfakeTIMIT	640	0	640	✅ HF-hosted
GenVideo-100K	188,169	100,000	88,169	✅ HF-hosted
HIDF	8,722	4,361	4,361	✅ HF-hosted
Hemgg/Open-Sora-Plan-v1.0.0	143,013	0	143,013	🔗 HF source
Hemgg/SDFVD-video-dataset	106	53	53	🔗 HF source
Hemgg/deep-fake-detection-dfd-entire-original-dataset	3,431	0	3,431	🔗 HF source
KhunPop/deepfake	3,296	1,349	1,947	🔗 HF source
SDFVD2.0	927	456	471	✅ HF-hosted
Sarim-Hash/video_DEEPFAKE_dataset	32	0	32	🔗 HF source
TonyStark03/deepfake	6,449	4,286	2,163	🔗 HF source
UADFV	98	49	49	✅ HF-hosted
UniDataPro/deepfake-videos-dataset	10	0	10	🔗 HF source
dfb-data/deep-fake-detection-cropped	3,293	0	3,293	🔗 HF source
hi-paris/FakeParts	81,732	28,523	53,209	🔗 HF source
lavandejoey/fake_temp	10,000	0	10,000	🔗 HF source
liusiyi641/deep_fake_videos	9,599	24	9,575	🔗 HF source
liusiyi641/deep_fake_videos_val	9,599	24	9,575	🔗 HF source
ohjoonhee/DF40-Fake-Videos	30,257	0	30,257	🔗 HF source
ud-biometrics/deepfake-dataset	10	0	10	🔗 HF source
xingjunm/WildDeepfake	7,314	3,805	3,509	🔗 HF source
TOTAL	913,830	339,405	574,425

Schema

Field	Type	Description
id	string	Globally unique ID
filepath	string	Original relative path
hf_path	string	Path in this HF repo (videos/…)
label	string	"real" or "fake"
split	string	"train" / "val" / "test" / "none"
dataset_source	string	Source dataset name
original_metadata	string (JSON)	Dataset-specific metadata

Citation

@misc{unified-deepfake-2024,
  title={Unified Deepfake Video Dataset},
  author={Your Name et al.},
  year={2024},
  publisher={HuggingFace},
  url={https://huggingface.co/datasets/belkhir-nacim/deepfake-videos}
}