Osen Mac-Iriase | 2026 I.S. Symposium
Name: Osen Mac-Iriase
Title: Doodle-To-Image: Enhancing Text-Guided Image Generation with Sketch Prompts in ComfyUI
Major: Computer Science
惭颈苍辞谤:听Education
Advisor: Daniel Palmer
How can we make AI image synthesis more predictable? This research investigates the impact of structural conditioning on generative AI by comparing text-only workflows with sketch-guided ones. Using Stable Cascade and ControlNet-style processing, simple doodles were used to guide the image-generation process. Experiments varying Classifier-Free Guidance (CFG) and seeds revealed that while text-only prompts often produce unpredictable layouts, sketch-based inputs lock in the composition. The study concludes that multimodal inputs are essential for achieving professional-level control in AI art tools.
Posted in Symposium 2026 on May 1, 2026.
