
Academic lab · Darmstadt, Germany
AIML LabTU Darmstadt
The AI and Machine Learning Lab at TU Darmstadt releases open safety tooling and reasoning research, including LlavaGuard and Safe Latent Diffusion.
Darmstadt, Germany
- Type
- Academic lab
- Based in
- Darmstadt, Germany
- Key papers
- 3
- Featured projects
- 8
About
The AIML Lab at TU Darmstadt is led by Professor Kristian Kersting, who joined the university in 2019 and also co-directs hessian.AI. Its stated aims are advancing statistical learning, probabilistic reasoning and collaborative intelligence, with work in statistical relational AI, generative modeling and trustworthy machine learning.
The lab publishes code on GitHub and models on Hugging Face, where it lists 20 models and 42 datasets. Safety tools include Safe Latent Diffusion, LlavaGuard and QwenGuard, and recent releases cover OlmoLogic reasoning models and the KletterMix German pretraining data.
Its audience is researchers and developers who need open checks for image-text datasets and generative models.
Mission and ambitionsThe lab aims to make machines learn about the world rapidly and flexibly, combining logical and probabilistic methods with deep learning. It emphasises trustworthy and safe generative systems.
Known for
- 01
Safe Latent Diffusion
- 02
LlavaGuard vision-language safeguards
- 03
Statistical relational AI
- 04
OlmoLogic reasoning models
Milestones
- 2022
Safe Latent Diffusion
Released a method and model for mitigating inappropriate image generation (arXiv 2211.05105).
- 2024
LlavaGuard
Released open vision-language safeguards for datasets and models (arXiv 2406.05113).
- 2025
QwenGuard
Published further image-text safety models alongside LlavaGuard on Hugging Face.
- 2026
OlmoLogic
Published reasoning models trained with RLVR and inductive logic programming.
- 2026
KletterMix
Released a German pretraining data collection.
Open source
Open-source repository. (209 stars)
A free, open-source and modern Chess Variant Analysis GUI for the 21st century (160 stars)
Rational Activation Functions - Replacing Padé Activation Units (107 stars)
Official Implementation of Safe Latent Diffusion for Text2Image (97 stars)
Model with 15 likes on the Hub.
Model with 11 likes on the Hub.
- AIML-TUDA/i2pdataset
Dataset with 21 likes on the Hub.
Space with 29 likes on the Hub.
Research papers
Notable releases
Openness
Models, datasets and code are public on Hugging Face and GitHub (ml-research).
People
- Kristian KerstingProfessor of AI and Machine Learning, head of the lab
Also heads a research department at DFKI.
- Patrick SchramowskiResearcher; lead author on Safe Latent Diffusion and LlavaGuard
- Manuel BrackResearcher; co-author of Safe Latent Diffusion and LlavaGuard
- Quentin DelfosseResearcher at TU Darmstadt
Listed from the lab's own pages and public profiles.
Contact and community
Contact via site.
Sources
Compiled from the lab's public pages, Hugging Face and GitHub. Last checked 2026-11-17. Something wrong or out of date? Tell us on Discord.



