Falcons AI's NSFW classifier ranks 6th on Hugging Face with 50.8M downloads
Falcons AI, an AI company based in Ras Al Khaimah, UAE, built an NSFW (not safe for work) image classification model that racked up 50.8 million downloads on Hugging Face in the 28 days to February 14, 2025. That download count puts it sixth by downloads among more than 1,400 models hosted on the platform.
By comparison, DeepSeek, one of the most talked about AI model families at the time, recorded 3.7 million downloads on Hugging Face over the same 28-day window, counting all its models combined. Falcons AI's single classifier logged 12 to 13 times more downloads than DeepSeek's entire lineup put together.
The model itself is a Fine-Tuned Vision Transformer (ViT), based on the transformer architecture commonly used in natural language processing but adapted for image classification. It was trained on 80,000 curated images to distinguish normal content from explicit content, using a learning rate of 5e-5 and a batch size of 16. It's released under the Apache 2.0 open-source licence, so it can be downloaded and used without payment.
Falcons AI, headquartered in Ras Al Khaimah, provides AI-powered solutions for businesses and governments, focusing on automation, efficiency and decision-making support, and has an open-source model programme; the article does not name any individual founders or executives, nor does it give revenue or funding figures for the company. The piece frames the download numbers as a sign that the UAE's investment in its AI ecosystem, spanning large research labs, commercial ventures and independent developers, is translating into global-scale adoption, not just headline model releases.
Key facts
- Falcons AI's NSFW image classifier passed 50.8 million downloads on Hugging Face in the 28 days to February 14, 2025.
- It ranks sixth by downloads among more than 1,400 models hosted on Hugging Face.
- Falcons AI's single model logged 12 to 13 times more downloads than all DeepSeek models combined (3.7 million) over the same period.
- The model is a Fine-Tuned Vision Transformer trained on 80,000 curated images, with a learning rate of 5e-5 and a batch size of 16.
- It's released under the Apache 2.0 licence by Falcons AI, a company based in Ras Al Khaimah, UAE.
Why it matters
Content moderation is a large and growing niche: platforms handling user-generated content, especially as generative AI tools flood the web with more images, need automated filters that scale. Falcons AI's download numbers show a narrow, practical tool outperforming headline-grabbing general-purpose models within that niche, on raw Hugging Face download counts. They also give the UAE's AI sector a visible global data point that comes from open-source distribution rather than from a marquee lab release.
Who it affects
Businesses and platforms building content moderation into their products, across media publishing, broadcasting, corporate governance and regulatory compliance, get a freely licensed classifier to draw on. Developers comparing open-source options on Hugging Face get a proof point for a specific vertical rather than for general chat or reasoning models. The story also affects how the UAE's AI sector is perceived, beyond well-known labs, by pointing to a commercial company's open-source model as evidence of broader ecosystem strength.
How to use it
The model is published on Hugging Face under Falcons AI's account and can be downloaded and used under the Apache 2.0 open-source licence, so it can be dropped into an image-classification or moderation pipeline without cost or a proprietary API. The article gives no pricing or paid tier, consistent with the model being free and open source.
How solid is it
The figures come from a single article on Middle East AI News, citing Hugging Face's own download counts as of February 14, 2025, a public, platform-reported metric rather than a company self-report. The comparison to DeepSeek's downloads over the same window, and the training details (80,000 images, the learning rate, the batch size), add specificity. The article does not name any individual at Falcons AI, gives no revenue or funding figures for the company, and does not explain how the classifier works beyond identifying it as a fine-tuned Vision Transformer.
Risks and caveats
Download counts are not the same as active or verified usage; the article does not say how many of the 50.8 million downloads reflect production deployments versus one-off testing, mirroring, or automated traffic. No accuracy, precision, or false-positive rate for the classifier is given, so its actual moderation quality is untested in the piece. The source is a UAE-focused outlet writing in a promotional register about the UAE's AI ecosystem, and it discloses nothing about Falcons AI's ownership or funding.