AI System Visualizes Dream Descriptions with Coherence.

Dreams, despite their emotional intensity, are often challenging to articulate and share effectively. To bridge this communication gap, researchers have developed the Dream Scene Visualiser (DSV) system. This innovative system takes a written description of a dream and transforms it into a chronological sequence of four distinct images, effectively creating a visual narrative of the dream experience. The process begins with a large language model (LLM) which is prompted to analyze the dream description and segment it into four chronological parts. Following this, a text-to-image model is employed to generate an image for each of these segmented parts. A crucial aspect of DSV is its focus on maintaining visual coherence across the entire four-panel sequence. If any generated image does not suitably match its corresponding text segment, the system is designed to regenerate that image until a satisfactory level of fidelity is achieved. The DSV system was evaluated using 50 visualisations derived from dream descriptions found in DreamBank. Objective measures, utilizing advanced vision-language models such as CLIP, DINOv2, and Qwen2-VL, were used to assess the quality, fidelity, and coherence of the generated visual sequences. The results demonstrate the system's capability to produce meaningful and visually consistent interpretations of complex dream narratives.