Unpaid Internship - AI - Stable diffusion & Generative AI Image generation - Berlin or remote

We are building Occibo(occibo dot com), an AI powered SaaS app that helps children’s book creators go from raw manuscript to print ready output fast. We are looking for a student intern(Unpaid mandatory Internship ) who wants hands on work in image generation, LoRA training, multimodal systems, and fast experimentation.

Remote.

Unpaid, study related internship. Mandatory internship as part of your degree is preferred. We also accept voluntary internship applications from non studying professional seeking professional expereince.

1. →Image generation experiments for children’s book illustration workflows, style consistency, character consistency, page by page art direction
2. →Dataset building and curation, reference collection, captioning, prompt sets, quality checks, bias and safety checks
3. →LoRA training and evaluation, style LoRAs, character LoRAs, control of pose, composition, and repeatable outputs
4. →Prompting systems and automated prompt generation, templates, constraints, negative prompts, reproducibility
5. →Multimodal agentic workflows using Langchain/Langgraph or any other agentic framework, linking manuscript context to visual briefs, retrieval of reference guides, image and text grounding
6. →Rapid testing harness, A, B comparisons, scoring rubrics, human review loops, failure analysis, iteration logs
7. →LLMOps and MLOps basics, experiment tracking, model versioning, evaluation reports, cost and latency tracking
8. →Building small prototypes that connect models to product flows, API wrappers, batch pipelines, internal tools for the team
9. →Documentation, clear notes, reusable notebooks, repeatable scripts, simple internal guides for future interns

1. Currently studying AI, ML, Data Science, or a related field
2. Can write Python code to test ideas quickly, iterate, and report results clearly
3. Curious, structured, and able to work independently with weekly check ins
4. Comfortable discussing tradeoffs like quality, cost, speed, and consistency

1. Any prior work with Stable Diffusion style models, LoRA, fine tuning, ControlNet, or similar
2. Text to Image - Midjourney prompt craft and systematic exploration