ThermVision: Exploring FLUX for Synthesizing Hyper-Realistic Thermal Face Data and Animations via Image to Video Translation

Muhammad Ali Farooq¹, Waseem Shariff¹, Peter Corcoran¹.

¹University of Galway

FLUX ThermVision LoRA Model Supplementary Materials Code ThermVision Sample Dataset ThermVision-DB Full Dataset Paper

Given a text-prompt into Thermally tuned Flux.1 Dev model...

Introduction

This work contributes to the field of image synthesis by introducing a novel diffusion-based framework, for generating high-quality synthetic thermal face images using the Flux diffusion model. By integrating a video retargeting pipeline, we address challenges in synthesizing diverse facial expressions, head poses, and subject-specific thermal characteristics. A key outcome is the creation of an open-source synthetic thermal face dataset, optimized for AI-driven applications such as thermal gender classification, facial detection, and landmark detection. Our approach enhances data availability in thermal imaging, reducing reliance on costly real-world data collection.

The work is accepted in ACM Multimedia Conference 2025 and will be presented at Dublin Royal Convention Centre Ireland.

Block Diagram

Text and video based conditioning is used to render thermal faces data with various smart transformations such as binary gender, male and female subjects with distinct facial headpose and expressions, facial thermal characteristic as demonstrated in below figure.

Intelligent transformation results on different subjects by using reference video conditioning and textual guidance prompts.

This dataset includes face data localization annotations to support tasks such as emotion recognition, gender classfication and pose estimation. We have provided precise bounding box annotations for each face in the images, as well as key face annotation labels, to enable researchers and developers to train and evaluate advanced computer vision models. These annotations are essential for tasks requiring accurate face localization and are designed to be easy to integrate into machine learning workflows.

54 Keypoints Facial Landmark Detection

We have rigorously evaluated the rendered data by running facial landmark detection tests using a pretrained thermal landmark detector. This evaluation ensures that the dataset is suitable for thermal-based facial landmark detection tasks. The pretrained detector has been applied to our dataset, and the results have been analyzed for accuracy, robustness, and consistency across different poses and thermal conditions. This evaluation provides confidence in the quality and reliability of the dataset for use in thermal vision applications.

Facial landmark detection on male subject with extreme headpose variation

Facial landmark detection on male subject with distinct facial expressions

Facial landmark detection on female subject with extreme headpose variation

Facial landmark detection on female subject with distinct facial expressions

Thermal Gender Classification and Facial Landmark Accuracy

This table shows classification accuracy for thermal gender classification across four benchmark datasets using three different deep learning models. Each model is evaluated under two training regimes: Purely Synthetic and Hybrid (Synthetic + Real).

BibTeX

@inproceedings{10.1145/3746027.3755448,
author = {Farooq, Muhammad Ali and Shariff, Waseem and Corcoran, Peter},
title = {ThermVision: Exploring FLUX for Synthesizing Hyper-Realistic Thermal Face Data and Animations via Image to Video Translation},
year = {2025},
isbn = {9798400720352},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
url = {https://doi.org/10.1145/3746027.3755448},
doi = {10.1145/3746027.3755448},
abstract = {High-quality thermal facial data is essential for advancing biometric recognition, surveillance, in-cabin driver monitoring, and human-computer interaction, all of which are integral for modern multimedia and interactive AI systems. In this work, we optimized the FLUX text-to-image diffusion model on diverse real-world thermal facial datasets to generate hyper-realistic 2D thermal facial images for both males and females, and propose a new dataset, ThermVision. To enhance their multimedia applicability, these images are processed through a video retargeting pipeline, where driving videos animate realistic facial expressions and head pose variations from a single 2D thermal image, producing high-fidelity thermal facial video sequences. The overall rendered dataset incorporates smart transformations, ensuring diversity across gender balance, extreme head pose variations, expressive facial dynamics, and facial accessories, making it a valuable resource for real-world applications. Additionally, we provide facial detection annotations to facilitate precise feature extraction and thermal-face analysis. To validate our synthetic dataset, we evaluate its effectiveness in thermal gender classification, as downstream machine learning task, along with thermal face localization and facial landmarks detection demonstrating its applicability in real-world scenarios. This approach significantly improves the availability, realism, and integration of thermal facial data, paving the way for more robust and immersive AI-powered thermal imaging applications. The dataset, code and associated models are available at- https://mali-farooq.github.io/ThermVision/},
booktitle = {Proceedings of the 33rd ACM International Conference on Multimedia},
pages = {10161–10170},
numpages = {10},
keywords = {diffusion models, flux, lora, lwir, synthetic data, text-to-image, thermal imaging},
location = {Dublin, Ireland},
series = {MM '25}
}

@article{FAROOQ2026112506,
title = {ThermVision-DB: A synthetic LWIR thermal face dataset for privacy-preserving thermal vision research},
journal = {Data in Brief},
volume = {65},
pages = {112506},
year = {2026},
issn = {2352-3409},
doi = {https://doi.org/10.1016/j.dib.2026.112506},
url = {https://www.sciencedirect.com/science/article/pii/S2352340926000594},
author = {Muhammad Ali Farooq and Waseem Shariff and Peter Corcoran},
keywords = {Synthetic data generation, Diffusion models, Thermal imaging, Image-to-video translation, Privacy-preserving AI, Multimodal facial analysis, Computer vision benchmark},
abstract = {ThermVision-DB presents a synthetic long-wave infrared (LWIR) facial dataset designed to support research in privacy-preserving vision, thermal perception, and multimodal facial analysis. The dataset builds upon generative diffusion models to create photorealistic thermal facial images and video sequences capturing controlled variations in facial expression and head pose. Each synthetic identity is generated using text-to-image conditioning followed by video retargeting module, enabling precise control over pose angles, expression intensity, and frame-to-frame consistency. The dataset includes a diverse set of synthetic adult identities of both male and female genders with multiple facial expressions - such as neutral, smile, frown, and surprise and head-pose rotations spanning yaw, pitch, and roll. Data are provided in both image and video formats, accompanied by face localization annotations, landmark detections and identity labels. To ensure reusability and scalability, all samples are generated through a standardized pipeline using open-source models, allowing researchers to easily expand the dataset with additional synthetic identities while maintaining consistent thermal appearance and scene illumination. The synthetic generation process avoids the use of any personally identifiable visual data, ensuring compliance with FAIR and GDPR principles. ThermVision-DB is intended for use in developing and benchmarking algorithms for facial detection, landmark localization, expression recognition, and head-pose estimation in the thermal domain. It also provides a foundation for research in synthetic-to-real transfer learning, privacy-safe biometric analysis, and cross-spectrum data fusion. The dataset is released for open research purposes under a non-commercial license, with full documentation and metadata available to facilitate reproducibility and integration with existing thermal vision benchmarks.}
}