GenTune: Toward Traceable Prompts to Improve Controllability of Image Refinement in Environment D...
summary
GenTune is a toolchain that generates structured prompts with LLM brainstorming, extracts labels with a label-extraction LLM, and links image regions to those labels so designers select a region, edit its label, and produce coherent refinements via either prompt-based regeneration or localized inpainting. A formative study with 10 designers informs two features: traceable prompt mapping and semantic-guided refinement; a within-subject evaluation with 20 designers shows significant gains in prompt–image comprehension, refinement quality, efficiency, and satisfaction (p < .01), and two studios adopt it in production; repeated identical elements remain hard to disambiguate.
AI-written summary · about
chapters (8)
- 0:00Introduction to GenTune
- 1:05Challenges in AI-assisted design
- 2:49Design goals of GenTune
- 3:11How GenTune works
- 4:57Evaluation and results
- 6:21Real-world deployment
- 7:44Future work and generalization
- 9:08Conclusion and Q&A
related talks
30mIs Prompt Engineering Dead? How Auto-Optimization is Changing the Game
18mTechniques for Enhanced Model Performance
29mIs Prompt Engineering Dead? How Auto-Optimization is Changing the Game
11mDesign Adjectives: A Framework for Interactive Model-Guided Exploration of Parameterized Design ...
30mIs Prompt Engineering Dead?
13mPrompt Engineering Simplified
27mQ&A #4
44mSupercharge Your Productivity Mastering Prompt Engineering