One Kiss: Emojis as Agents of Genre Flux in Generative Comics

Abstract: Generative AI has made visual storytelling widely accessible, yet current prompt-based interactions often force users into a trade-off between precise control and creative flow. We present One Kiss, a co-creative comic generation system that introduces "Affective Steering". Instead of writing text prompts, users guide the tone of their story through emoji inputs, whose semantic ambiguity becomes a resource rather than a limitation. Unlike traditional text-to-image tools that rely on explicit descriptions, One Kiss uses a dual-stream input in which users define structural pacing by sketching panel frames and set atmospheric tone by pairing keywords with emojis. This mechanism enables "Genre Flux," where emotional inputs accumulate across panels and gradually shift the genre of a story. A preliminary study (N = 6) suggests that this soft steering approach may reframe the user's role from prompt engineer to narrative director, with ambiguity serving as a source of creative surprise rather than a loss of control.

• The paper introduces One Kiss, a co-creative comic system that combines sketch-based spatial pacing with emoji-keyword pairings and a decay-based four-dimensional state vector to steer genre across panels.
• In an exploratory study of six participants, all triggered a genre shift by panel four, while users reported directing mood and narrative pacing rather than debugging technical prompts.
• The findings suggest ambiguity can encourage serendipitous story development, but persistent keywords, repetitive close-ups, single-character anchoring, and limited evaluation constrain the system’s current generalizability.

Motivation and design stance

The paper addresses a recognized weakness in prompt-based generative interfaces: text prompts are effective for specifying objects but poorly suited to conveying implicit affective qualities, and the iterative correction they demand resembles debugging rather than creative expression (Schick et al., 2023). Rather than improving prompting precision, the authors invert the premise by treating ambiguity as a deliberate design resource, drawing on Gaver et al.'s framework of ambiguity in design. The chosen medium is the emoji, which prior work shows carries quantifiable emotional valence while remaining semantically open to interpretation.

System architecture

One Kiss decouples narrative control into two parallel input streams that converge on a Stable Diffusion backend.

Spatial Pacing stream: users sketch bounding boxes per panel; box aspect ratio acts as a structural prompt for composition (wide boxes bias toward cinematic panoramas, tall narrow boxes toward close-ups), letting users control "camera language" without technical vocabulary.

Affective Injection stream: users pair interface keywords (e.g., "Gunshot") with emojis (e.g., "Wilted Flower"), and these pairings drive tone. A color-shifting particle system provides feedback on system state.

Genre Flux Algorithm: the core contribution over stateless generation. The system maintains a Narrative State Vector $S_t$ over four dimensions — Romance, Tragedy, Chaos, Mystery — updated as

$$S_{t} = S_{t-1} \cdot \alpha + (W_{kw} + W_{emoji}) \cdot \beta$$

with decay factor $\alpha = 0.8$ (empirically tuned) and a rarity multiplier $\beta \in [1.0, 3.0]$ derived from keyword semantic rarity, under the assumption that unusual inputs signal intentional pivots. When a dimension crosses threshold $\theta$, Style Modifiers inject genre constraints into both positive and negative prompts (e.g., "monochrome blue palette, film noir grain" with suppressed "cheerful expressions"). A Global Character Anchor prepends a fixed protagonist description to every request to prevent identity drift while style drifts intentionally.

Two scope limitations should be noted at this point: the four-dimensional emotion space is explicitly not presented as a fixed taxonomy but a prototype choice, and the character anchoring supports only single-protagonist narratives.

Preliminary evaluation

Six participants (ages 22–26, mixed generative-AI experience) completed a think-aloud remote study: tutorial, then creation of a six-panel comic drifting from neutral tone into a target genre, followed by semi-structured interviews.

Two authors coded all 36 generated panels against stated target genres using palette, lighting contrast, and compositional density; all six participants successfully triggered a Genre Flux by the fourth panel, with disagreements resolved through discussion. Qualitatively:

• Role reframing: four of six participants described the interaction as directing rather than prompting. One participant characterized pairing a keyword with an emoji as "giving an actor a mood cue rather than writing a technical script."
• Serendipity over frustration: ambiguity was embraced rather than experienced as control loss. One participant's "Clown" + "Rain" pairing produced an unanticipated surreal background the user adopted as a plot twist; another steered an entire thriller arc ending in a kiss that was unplanned yet perceived as coherent — evidence that Genre Flux can produce sequence-level, not merely panel-level, surprises.
• Perceived agency: users traded pixel-level control for pacing and genre influence, attributing their sense of agency to originating structural decisions themselves.
• Reported tensions: early keyword choices anchored later panels to initial settings ("Coffee Shop" persisting past its narrative relevance), and consecutive panels defaulted to similar close-up compositions — indicating the spatial stream lacks the gradual evolution the affective stream supports.

Given $N=6$, the authors explicitly frame these findings as exploratory; no quantitative baseline comparison or validated instrument was used.

Discussion

The paper positions the work against structural steering systems (TaleBrush's fortune curves, ID.8's dialogue planning, Opal's multimodal cues) and sequential visual story generators (StoryDALL-E), arguing that none allow affective signals to accumulate and reshape genre across a sequence — each panel remains tonally independent. One Kiss fills that specific gap. The broader claim advanced is a shift from lexical precision toward affective resonance, with emoji acting as semantic scaffolds for creative exploration. The central unresolved design tension is between intentional stylistic destabilization and coherence: the authors acknowledge the balance currently depends on the fixed Character Anchor, and propose allowing users to lock visual elements immune to flux.

Limitations and open questions

The evidence base is thin by design: six participants, one task format, self-reported perceptions, and subjective two-author coding of genre shift. Key open questions include whether the effect generalizes beyond single-protagonist stories, how the four-dimension state space scales, whether spatial pacing can support gradual evolution analogous to Genre Flux, and how thresholds ($\alpha$, $\theta$, $\beta$) generalize across user populations given their empirical tuning during pilots.

Conclusion

One Kiss contributes a prototype demonstrating that deliberately imprecise emoji inputs, accumulated through a decay-based state mechanism, can steer genre in generated comics without conventional prompting. Its preliminary study suggests soft control may reduce blank-page anxiety and redistribute agency from rendering to direction, while conceding that spatial steering has not achieved comparable flexibility.