Beyond Generation: An Empirical Study on Redefining the Act of Drawing Through an 85% Time Reduction in Picture-Book Production

Abstract

Conventional picture-book production imposes substantial physical and temporal demands on creators, often constraining opportunities for high-level artistic exploration. While generative AI can drastically accelerate image generation, concerns remain regarding style homogenization and the erosion of authorial agency in professional practice. This study presents an empirical evaluation of an AI-collaborative workflow through the full production of one professional 15-illustration picture-book title, and compares the process with a conventional hand-drawn pipeline by the same creator. Quantitatively, the proposed workflow reduces total production time by 85.2% (from 2,162.8 to 320.4 hours), with the largest substitution observed in early drafting stages. Qualitatively, however, the core contribution is the strategic reallocation of labor: time saved in mechanical rendering is reinvested into high-level Judgment (aesthetic selection, narrative direction, and cross-scene consistency decisions) and Completion (embodied manual retouching and integrative refinement). Notably, 235 hours were devoted to Completion, indicating that publication-quality outcomes still depend on sustained human synthesis to reconcile generative inconsistencies. Our findings suggest that AI-integration, when framed as a "mild-work" partnership, enhances rather than diminishes the creative experience by shifting the creator's focus from repetitive physical labor to sophisticated aesthetic synthesis.

Figures (7)

• Figure 1: Comparison of conventional and AI-collaborative picture-book illustration workflows. (a) Carpe Diem, produced with a conventional hand-drawn pipeline (1,203.6 hours for illustration). (b) Golden Drops Opening the Sky, produced with our AI-collaborative workflow (1.2 hours for initial AI output + 235.3 hours for human Completion). The proposed method inherits and further develops picture-book-specific stylistic signatures from the author’s prior works. More importantly, by reinvesting recovered time into aesthetic selection and scene direction, the workflow yields richer narrative cues in props and architecture, as well as a stronger sense of atmospheric presence. Reported times indicate total effort per one picture-book title (15 illustrations). All illustrations © Cosei Kawa. All rights reserved.
• Figure 2: Comparison of production time breakdown between conventional and proposed workflows. While total time is reduced by 85%, the proportion of time dedicated to 'Judgment & Completion' significantly increases in the proposed method.
• Figure 3: System architecture workflow of our human--AI collaborative picture-book pipeline. The architecture operationalizes role separation: AI handles rapid provisional Output, while the creator reinvests recovered time in Judgment (aesthetic selection) and Completion (manual final integration). Dashed module boundaries indicate technical layers that guarantee this division of labor across UI-level parameter engineering, backend asynchronous generation, and automated metadata management.
• Figure 4: Graphical User Interface of the 'Prompt Recipe' system. Note: The UI is in Japanese as it was natively developed for the author's production environment. (A: Style Slider, B: Prompt Queue for iterative generation based on the author's legacy assets.)
• Figure 5: Total production time comparison per picture book (15 illustrations). The conventional hand-drawn pipeline is compared against the actual measured time of the AI-collaborative pipeline.