Multimedia authoring has historically depended on explicit manipulation paradigms such as layers, masks, timelines, Bezier paths, rigging systems, and non-destructive editing graphs, all of which give creators precise local control but demand substantial technical literacy and time. The post-2020 generative turn has reconfigured this workflow by making natural language, sketches, depth maps, and sparse geometric constraints viable interfaces to image and video synthesis. Rombach et al. (2022) established latent diffusion as a computationally tractable foundation for high-quality visual generation, Zhang et al. (2023) extended that foundation with ControlNet so that prompts could be anchored to edges, poses, and segmentation maps, and Cai et al. (2024) showed that even video generation becomes more authorable when low-fidelity animated meshes are fused with pretrained diffusion models. Systems such as Adobe Firefly and related commercial tools have operationalized this shift for practitioners, but the underlying research tension remains unresolved: prompt-based authoring is semantically rich yet procedurally opaque, whereas traditional multimedia tools are procedurally explicit yet semantically laborious. In practice, this means AI authoring systems are strongest during ideation, style exploration, and rapid variation, but they remain weak at revision locality, provenance tracking, and preserving an author's exact intent across iterative edits. A critical research gap is the absence of a unified representation that links prompts, user constraints, scene structure, edit history, and asset identity into a reversible authoring graph rather than a sequence of loosely related generations. Future research should therefore treat multimedia authoring not as a one-shot generation problem but as a longitudinal interaction problem requiring controllability, memory of prior edits, and interoperable metadata. The real frontier is not merely generating attractive outputs from text, but building mixed-initiative systems in which human intentionality remains first-class and every AI-mediated change can be audited, refined, and compositionally reused across downstream media production (Rombach et al., 2022; Zhang et al., 2023; Cai et al., 2024).
References1. Rombach, R., Blattmann, A., Lorenz, D., Esser, P., & Ommer, B. (2022). High-resolution image synthesis with latent diffusion models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR 2022).
2. Zhang, L., Rao, A., & Agrawala, M. (2023). Adding conditional control to text-to-image diffusion models. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV 2023).
3. Cai, S., Ceylan, D., Gadelha, M., Huang, C.-H. P., Wang, T. Y., & Wetzstein, G. (2024). Generative rendering: Controllable 4D-guided video generation with 2D diffusion models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR 2024).
S. M. Monowar KayserLecturer, Department of Multimedia & Creative Technology (MCT)
Faculty of Science & Information Technology
Daffodil International University (DIU)
Daffodil Smart City, Savar, Dhaka, Bangladesh
Visit: https://monowarkayser.com/
