TaleBrush: Sketching Stories with Generative Pretrained Language Models

DramatVis Personae: Visual Text Analytics for Identifying Social Biases in Creative Writing

“…Many tools in this area generate stories iteratively: a writer provides an initial prompt for the story; AI generates a section of the story automatically; the writer edits the generated story and provides further prompts and so on [12,13,54]. Recently, Chung et al [11] proposed TaleBrush, an interactive tool where AI can generate a narrative arc based on a sketch drawn by a writer that outlines the expected changes in the fortunes of characters. Other works include IntroAssist [43], a web-based tool that helps entrepreneurs to write introductory help-requests to potential clients, investors, and stakeholders; INJECT [59], a tool for supporting journalists explore new creative angles for their stories under development; and an interactive tool proposed by Sterman et al [87] that allows writers to interact with literary style of an article.…”

From Tool to Companion: Storywriters Want AI Writers to Respect Their Personal Values and Writing Strategies

“…In [38], the collaborative tool supports metaphor creation based on the writer's context. Breaking away from suggestion-based co-ideation approaches, visual line sketch interaction has recently been used as a novel AI steering mechanism for a GPT-based language model in [20], giving writers an intuitive understanding of the AI's output and allowing them to control the protagonist's fortune while collaboratively generating a narrative outline. Beyond ideation, tools such as [93], can provide translating support by ofering sentence-level suggestions, although they aford limited control over suggestion topics and direction, sometimes deviating from the narrative [16].…”

“…However, ofering users a useful explanation is challenging due to the complexity of large language models. The ideation tool in [20] takes a decisive step in addressing both these challenges in the feld of AI co-writing by providing sketching control to guide characters' fortune in the narrative and a visualization of the model's output to support sensemaking. At the same time, this sketch-based interaction approach intuitively communicates to users that there will be some degree of uncertainty in the output text, lowering their expectations for control.…”

“…For example, Microsoft Word, Grammarly, and ProWritingAid are all popular solutions for supporting syntactic and semantic revision (i.e., grammar, spelling, and style checking). Considering ideation, this area has been explored extensively in creative writing [16,20,21,38,39,81,92]. While there has also been some focus on translating support in recent years [2,23,93,97], the current work on machine-assisted storywriting ultimately misses out in investigating writers' needs for AI-led writing support in the translating process, revealing a crucial gap.…”

Loose Ends: A Mixed-Initiative Creative Interface for Playful Storytelling

“…One recent mixed-initiative storytelling support system that departs from the interaction paradigm of local continuation suggestion is TaleBrush (Chung et al 2022), which instead aims to give users direct control of high-level story structure via the sketching of a visual fortune arc for the story's main character. This approach has so far only been used to generate very short stories (on the order of five sentences long), and the coherence of the generated stories is limited, but this potential alternative means of specifying high-level storytelling goals still merits mention here.…”