Internship Project
Start-Ups and Entrepreneurship

recreategoods – AI-Driven Image Editing and Diffusion Models for Sustainable Design

Institution
Humboldt-Universität zu Berlin
Humboldt Innovation
Subject Area
Artificial Intelligence · Computer Vision · Generative Models · Machine Learning · Creative Computing · Sustainable Fashion Technology
Availability
04 May – 24 July
18 May – 07 August
01 June – 21 August
15 June – 04 September
 
Internship Modality:
On-site internship in Berlin

Applications for summer 2026 are open from 05 November to 18 December 2025.
Project Supervisor(s)
Thomas Chille
Academic Level
Advanced undergraduate students (from second year) 
Master's students 
Ph.D. students 
Language
English
Further Information
Project Type
Start-Up
Project Content
The project explores AI-driven image editing and diffusion models in the context of industrial upcycling and fashion design.
Our goal is to adapt and integrate state-of-the-art open-source architectures—such as Flux.1 Kontext and Qwen3 Image—into multi-model, agentic AI systems that support creative and sustainable design workflows.
A central focus is the development of a domain-specific Image Edit Benchmark that systematically evaluates model performance and visual quality across design-relevant editing tasks.
The research connects machine learning, computer vision, and computational creativity, bridging academic insight with real-world applications in the creative industries.
Tasks for Interns
Interns will work at the intersection of AI research and creative technology, contributing to the design and implementation of an Image Edit Benchmark and related model integration pipelines.
Tasks may include:
  • Curating and annotating benchmark datasets
  • Developing evaluation code and metrics for model comparison
  • Integrating diffusion or editing models (Flux.1, Qwen3, etc.) into prototype pipelines
  • Running controlled experiments to assess and improve model performance
  • Documenting findings and contributing to potential co-authored publications

Depending on academic level, interns may extend their work into a Bachelor’s, Master’s, or PhD-related thesis, with mentoring from both the startup and academic partners.

Academic Level
Advanced undergraduate students (from second year) 
Master's students 
Ph.D. students 
Requirements
  • Strong interest in image generation, diffusion, and editing models
  • Solid programming experience in Python (PyTorch or similar frameworks)
  • Basic understanding of deep learning architectures and training workflows
  • Ability to work with datasets, benchmarks, and model evaluation tools
  • Interest in open-source research and interdisciplinary AI applications
  • Independent, analytical, and collaborative working style

Experience with image editing pipelines, LoRA fine-tuning, or evaluation metrics (FID, CLIPScore, etc.) is a plus but not mandatory.

Expected Preparation
Interns should familiarize themselves with:
  • Core papers and repositories on image diffusion (e.g., Stable Diffusion, Flux.1, Qwen-Image)
  • Evaluation methods for image editing quality and consistency
  • Basic tools for dataset management (e.g., Hugging Face Datasets, COCO format, Parquet)
  • Principles of agentic AI systems and multi-model orchestration

Prior coursework or self-study in deep learning, computer vision, or AI for creative applications is highly recommended.

Back to Project List

For more information on the Humboldt Internship Program or the project, please contact the program coordinator.