Request access to ALADDIN
ALADDIN is released for non-commercial research only, under the ALADDIN License and the licenses of its third-party components.
ALADDIN is a face-swapping model. Access is granted for non-commercial research only. Before requesting access, read the ALADDIN License (LICENSE) and the licenses of the third-party components listed in this model card. By requesting access, you agree to all of them. Access may be revoked if they are violated.
Log in or Sign Up to review the conditions and access this model content.
ALADDIN (gating test)
This is a test repository for the access and license settings. It contains no model weights.
ALADDIN (Adversarial LAtent Diffusion Distillation for One-Step Identity SwappiNg) is a one-step face-swapping model, accepted to WACV 2027.
- Paper: https://arxiv.org/abs/2610.37503
- Code: https://github.com/isdevnull/aladdin
- Detector for ALADDIN-generated images: https://huggingface.co/Sumsub/aladdin-detector
Intended use
ALADDIN is released for non-commercial research, in particular research on face generation and on the detection of face swaps.
Out-of-scope use
ALADDIN must not be used to impersonate real people, to commit or prepare fraud or identity theft, to attack or bypass identity verification, KYC or liveness systems without the written authorization of their operator, or to create content depicting a real person without their consent, including sexual, defamatory or harassing content. The full list of restrictions is in the ALADDIN License.
Responsible release
Face swapping can be misused to produce non-consensual or deceptive imagery, and a one-step method lowers the cost of doing so at scale. Our interest in the problem comes from identity verification, where realistic swaps are needed to train and stress-test the systems that detect them. Publicly available tools already produce convincing swaps, so defenders need training examples at least as strong as those available to attackers.
We therefore release ALADDIN in stages. First, we openly release a classifier that identifies images generated by ALADDIN. It is specific to our model and is not a general deepfake detector, and detection alone should be combined with other safeguards. The code is publicly available, while the model weights are available only through gated access for non-commercial research.
License
The ALADDIN Model is licensed by the ALADDIN authors under the ALADDIN License, which consists of the AI Pubs Open RAIL-M License v0.1 and the ALADDIN Additional Terms. The ALADDIN License allows non-commercial research use only and prohibits, among other things, illegal use, harmful or abusive content, impersonation, and attacks on identity verification systems. It does not permit redistribution of the weights.
Third-party components
ALADDIN is built on, and is used together with, the third-party models below, and it was trained with the help of third-party models and datasets. These components remain subject to their own licenses, and any use of them, including any manual or automatic download, must follow those licenses. Several of them allow non-commercial research use only. Nothing in the ALADDIN License grants rights in these components. To the extent that there is any conflict between the ALADDIN License and a third-party license, the more restrictive terms apply.
Required to run ALADDIN (download them from their original sources; their licenses apply to your use): FLUX.1 Krea [dev], PuLID, EVA-CLIP, InsightFace face recognition models, OminiControl.
Used to train ALADDIN: Arc2Face dataset, VGGFace2-HQ, DECA and FLAME, 6DRepNet, SegFormer face-parsing model, LPIPS, improved aesthetic predictor.
FLUX.1 Krea [dev] attribution. The FLUX.1 [dev] Model is licensed by Black Forest Labs Inc. under the FLUX.1 [dev] Non-Commercial License. Copyright Black Forest Labs Inc. IN NO EVENT SHALL BLACK FOREST LABS INC. BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH USE OF THIS MODEL. The ALADDIN authors have modified the FLUX.1 [dev] Model by training LoRA adapters and identity-conditioning layers on top of it. ALADDIN is not an official product of Black Forest Labs Inc. and has not been endorsed by it.
Citation
@misc{shchekotov2026aladdinadversariallatentdiffusion,
title={ALADDIN: Adversarial Latent Diffusion Distillation for One-Step Identity Swapping},
author={Ivan Shchekotov and Timofei Gritsaev and Viacheslav Pirogov and Dmitry Vetrov},
year={2026},
eprint={2610.37503},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2610.37503},
}
Model tree for Sumsub/aladdin-gating-test
Base model
black-forest-labs/FLUX.1-dev