Imagination-to-Image: Intended Visualisation of Imaginations with AI
How can AI turn a person’s inner imagination into a picture that stays faithful to the intended idea? This project wants to answer this question.
Duration: July 2024 – December 2024
Status: Completed
Educational Level: Lower Secondary Level, Upper Secondary Level – Vocational Education, Upper Secondary Level – Grammar School Education, Tertiary Level
Topic: Artificial Intelligence AI, Digital Skills & Literacy
Keywords: Digital Skills, Effective Learning Habits, Artificial Intelligence, Metacognition, Problem Solving, Process Data
Initial Situation
How can AI image tools turn a person’s inner imagination into a picture that stays faithful to the intended idea (not just AI’s own creativity), and when is this useful in education and design processes? The project investigates “closeness” between the original imagination and the AI output as a success criterion.
Objectives
- Assess suitability of AI generators for visualising imaginations, using a criteria catalog to judge closeness to the original idea.
- Develop and optimize prompting (“how to talk about images”), leveraging HKB expertise in visual language and perception.
- Implement and apply results in practice (e.g., teaching-learning formats, makerspaces, brainstorming, strategy documentation).
Method
- Test series with multiple AI generators, evaluated via the criteria catalog for “closeness.”
- Prompt engineering experiments grounded in image description and interpretation principles.
- Hands-on implementations (e.g., education futures visualisations), exploring workflows like sketch+text+inpainting and digital post-editing.
Results
- AI can create impressive images but often misses exact intent; pure text-to-image struggles with complex scenes.
- Best results came from combining sketch+text+AI processing (incl. inpainting) plus manual/digital post-editing; breaking complex scenarios into parts and composing a collage helps.
- Ethical issues surfaced (e.g., gender bias in outputs) and the need for careful parameter control and prompt design.
Implemented Translation
Impact on educational practice:
- Idea-to-visual support for planning & teaching.
- Better stakeholder alignment: “Intent-closeness” criteria help teams judge whether an image truly represents the original idea, improving common understanding in curriculum and space-of-learning projects.
- Ethically aware adoption. The project surfaces bias risks (e.g., gendered depictions) and shows how prompt/parameter control mitigates them—raising teachers’ critical AI literacy.
The project delivers criteria, prompts, and workflows schools and teams can use to externalize ideas for teaching-learning formats, makerspaces, brainstorming, and strategy work, enabling clearer stakeholder communication about future concepts of learning and schooling. Outputs include a final report and a video presentation; both can be found on the project website.
