The open-source engine
Hotdog / Not Hotdog
Apache-2.0 · CLIP-based · Self-hosted · Yes, we named it on purpose.
What it is
An image and video classifier that flags NSFW content and runs locally. Built on CLIP. Supports batch processing, GPU acceleration, and video frame sampling. Untrained by default, you bring a rules-based policy or train your own classifier on top, which means you control what the thing actually looks for.
Useful for ingest pipelines that need a local safety layer without calling out to a commercial API, and for teams that want to stay in control of the classification logic itself rather than delegate it to a vendor black box.
What you get
- Local inference. No calls out, no accounts, no rate limits.
- CLIP backbone. Same primitives used across modern vision-language classifiers.
- Batch + GPU. Designed for pipelines, not interactive demos.
- Video frame sampling. Classify video without re-encoding or per-frame glue code.
- Bring your own policy. Rules-based or trained head on top of the CLIP embedding.
- Apache-2.0. Use it commercially. Ship it in products. Fork it.
Install
Clone the repository and follow the README. Python, CUDA-optional.
The engine is a tool. The service is a gateway.
Hotdog / Not Hotdog is a classifier you run yourself. SFWaaS is a private-deployment moderation gateway operated by us, with a verified buffer, chain-of-custody break, multimodal ingest, operational support, and a contract.
One is a tool you own. The other is infrastructure you subscribe to. They solve adjacent problems. Neither is a lesser version of the other.
If you get the reference, you understand the product. If you don’t, the product still works.
Using it in production
A few teams already run the engine at meaningful volume. Whether you should run it yourself or use the SFWaaS gateway depends on whether you want to own the operational surface, throughput scaling, failure handling, policy drift, modality coverage, audit logging, or hand it off. There is no correct answer; there is a correct answer for your situation.
If you want to talk about the operated version instead: [email protected].
Contributing
Issues and pull requests on GitHub. Contributions under Apache-2.0. Please run the existing test suite before opening a PR, the engine is used in production by a handful of Creative Mayhem products, so regressions matter.