Persode: Personalized Visual Journaling System

• Persode is a personalized digital journaling system that integrates reflective conversation and visual storytelling to produce diary text entries and personalized illustrations.
• It uses a retrieval-augmented framework to prioritize emotionally significant memories and adapts outputs based on demographic and stylistic preferences.
• The system streamlines journaling for digital-native users by offering engaging, fast-paced, and interactive visual feedback aligned with modern digital consumption habits.

Persode is a personalized visual journaling system that combines personalized onboarding, a memory-aware conversational agent, and automated visual storytelling to convert reflective conversation into a diary text entry and a personalized illustration. It is designed for Generation Alpha and Generation Z, who are described as preferring visual, fast-paced, interactive, and socially familiar digital experiences over traditional text-heavy journaling. The system uses a Retrieval-Augmented Generation framework to prioritize emotionally significant memories, ground responses in user history, and adapt both textual and visual outputs to demographic, stylistic, and conversational preferences (Jin et al., 28 Aug 2025).

1. Problem setting and design motivation

Persode addresses a mismatch between the established benefits of reflective journaling and the interaction preferences attributed to digital-native users. Reflective journaling is framed as valuable for self-awareness, emotional regulation, and well-being, yet traditional journaling formats are described as too text-heavy and slow, lacking immediate feedback, offering low visual engagement, providing limited personalization, exhibiting weak emotional resonance, and often failing to maintain continuity across sessions because they do not remember past interactions well enough to support a meaningful long-term reflective relationship (Jin et al., 28 Aug 2025).

Within this formulation, the central design problem is not merely diary generation. It is the construction of a journaling environment in which personalization, responsiveness, and visual immersion are treated as primary system requirements. Persode therefore aims to bridge the gap between traditional reflective journaling and the personalized, visually immersive digital experiences associated with Generation Alpha and Generation Z.

This framing also narrows the target artifact. Persode does not produce only free-form conversation and does not operate only as a text diary assistant. Its intended output is a paired artifact consisting of a reflective diary text entry and a personalized illustration, with both components conditioned on user identity, user style, and remembered prior experiences.

2. System architecture and interaction flow

The full system comprises three major components: personalized onboarding, a memory-aware conversational agent, and automated visual storytelling. These components are organized into a pipeline in which users first specify personalization variables, then converse with the system about daily events and emotions, and finally receive a text diary entry and image generated from the conversational record (Jin et al., 28 Aug 2025).

Component	Main inputs	Role
Personalized onboarding	demographic and stylistic preferences; chatbot personality	stores preferences used later in text and image generation
Memory-aware conversational agent	user dialogue; extracted event-emotion metadata	maintains continuity and context-aware responses
Automated visual storytelling	memory metadata; onboarding preferences	generates a diary text and a personalized illustration

The onboarding stage captures information such as age, persona appearance, hair color, glasses, fashion style, background aesthetics, visual style, and chatbot personality traits such as empathetic, friendly, detailed, or direct. These preferences are stored in a database and later reused to shape chatbot responses, guide memory retrieval, frame reflective summaries, and construct image-generation prompts.

The conversational stage begins with user dialogue entering a temporary buffer. An Event-Emotion Analyzer then extracts structured memory units containing an event, an emotion, a timestamp, and hashtags or contextual tags. These units are stored as episodic fragments. A Memory Strength Scoring Mechanism ranks the fragments, and a Memory Selection Block retrieves relevant past experiences so that the chatbot can respond using prior context.

The generation stage converts the conversation into two outputs. Persode uses GPT-4o to generate a concise text diary entry and DALL·E 3 to generate an accompanying illustration. The result is intended to be emotionally coherent, visually appealing, and aligned with the user’s identity and stylistic preferences.

3. Episodic memory model and Retrieval-Augmented Generation

Persode’s memory subsystem is organized around episodic memory rather than undifferentiated chat history. A memory unit is defined as an event associated with an emotion, and longer dialogues are segmented into discrete events whose fragments retain metadata such as event-emotion pairs, timestamps, hashtags, and contextual information. The Retrieval-Augmented Generation component uses these stored fragments to retrieve relevant past experiences, preserve context across sessions, ground responses in user history, and avoid generic replies (Jin et al., 28 Aug 2025).

The paper specifies a memory strength score:

$$S \;=\; d(\Delta t)\,\cdot\, \frac{w_E\,E \;+\; w_R\,R \;+\; w_C\,C}{\,w_E \;+\; w_R \;+\; w_C\,}$$

where $S$ is the memory strength score, $E$ is emotional intensity, $R$ is recall frequency, $C$ is contextual relevance, $w_E, w_R, w_C$ are tunable weights, $d(\Delta t)$ is a time-based decay function, and $\Delta t$ is the elapsed time since the memory was formed. A possible decay function is also given:

$d(\Delta t)=e^{-\lambda \Delta t}$

This formulation ranks memories by combining emotional salience, repeated recall, contextual relevance, and temporal recency. The accompanying explanation states that recent memories matter more, but older memories can remain strong if they are emotionally important or often revisited. If the decay is ignored or absorbed into another term, the score reduces to a normalized weighted average of the three factors.

The design rationale is explicitly linked to Ebbinghaus’ Forgetting Curve, Levels of Processing Theory, and prior long-term memory systems like MemoryBank. The intended effect is a memory module that approximates human-like forgetting: recent memories are easier to retrieve, emotionally intense memories persist longer, and less important memories can fade to reduce computational load.

4. Personalization as a system-level principle

Persode treats personalization as a property of the entire journaling workflow rather than a post hoc stylistic adjustment. The onboarding process captures demographic and persona-related traits, environmental preferences, conversational style, and narrative tone. The listed preference categories include age, appearance attributes, hair color, glasses, fashion style, background theme, aesthetics, empathetic or direct response style, and reflective versus factual framing (Jin et al., 28 Aug 2025).

These preferences influence four distinct processes. First, they determine chatbot personality, so response tone adapts to the selected style. Second, they affect memory interpretation, since emotionally relevant memories are prioritized. Third, they condition text diary generation, with GPT-4o guided to write in a style aligned with the user. Fourth, they condition image generation, with DALL·E 3 prompts encoding appearance, fashion, and background preferences.

This architecture implies that personalization is distributed across representation, retrieval, and generation. It is not limited to surface vocabulary choices. User identity and style shape how memories are selected, how events are framed, how reflection is verbalized, and how the illustrated diary is visually rendered. The paper presents this as a response to the limitations of rigid templates and generic prompts in standard journaling tools.

5. Visual storytelling and illustrated diary generation

Persode’s visual storytelling module converts conversational reflections into an illustrated diary through a prompt-construction pipeline. The system extracts emotional and event metadata from the conversation, combines that information with onboarding preferences, and uses a Few-Shot Template System to create structured prompts for both diary text generation and image generation (Jin et al., 28 Aug 2025).

The prompt constructor takes as input extracted event details, emotion labels, contextual hashtags, and onboarding preferences such as age, fashion, glasses or hair, background style, and narrative tone. The resulting prompts are adapted along four named dimensions: characters, backgrounds, art style, and emotional tone. Character adaptation includes appearance, age, hairstyle, clothing, and accessories. Background adaptation covers scenery, setting, and atmosphere. Art style is aligned with user preference, while emotional tone may be sorrowful, reflective, joyful, or contemplative.

Two examples in the paper illustrate the intended behavior. In one, a teenage girl with dyed yellow hair is scolded by her mother and feels regretful; the generated prompt is described as preserving character identity, emotional state, visual setting, and personalized style choices. In another, a car splashes water on the user and ruins a favorite outfit; extracted tags include #FavoriteOutfit, #Upset, and #Laundry, and the resulting image may depict a contemplative character washing clothes, visually echoing the emotional event.

The stated objective is that the image not function as a generic illustration but as a visual reflection of the user’s own memory and style. This suggests a multimodal interpretation of journaling in which reflection is externalized simultaneously as narrative text and as a stylized scene grounded in remembered events.

6. Evaluation status, limitations, and adjacent research context

Persode is presented primarily as a system or design paper rather than a completed evaluation paper. The reported empirical evidence does not include a full formal user study with participant numbers, quantitative metrics, controlled comparisons, or statistical tests. What is reported is that diary generation takes about 3 seconds on average, and the paper claims that the design should improve engagement, emotional resonance, personalization, reflective depth, and continuity of memory (Jin et al., 28 Aug 2025).

The limitations are explicit. The paper acknowledges the absence of extensive user testing, the need to validate whether visual journaling strengthens emotional connection, possible variability in personalization quality across users, unknown long-term effects on engagement and satisfaction, and open questions concerning scalability and adaptation across demographics. Proposed future work includes extensive user testing, real-world deployment, longitudinal studies, and examination of usability, performance, emotional impact, and engagement over time.

The paper also frames Persode comparatively against traditional journaling, existing tools such as Reflectly, Reflectr, and Draw My Day, and other AI journaling systems that rely mainly on text or templates. A plausible implication is that Persode extends several strands of personalization research into the journaling domain. Persona-grounded dialogue work has emphasized coherence, personalization, and instruction adherence in conversational generation (Afzoon et al., 4 Feb 2026); mutual persona perception has framed dialogue quality partly in terms of whether interlocutors can infer one another’s personas (Liu et al., 2020); and speaker persona detection has treated persona inference from conversational text as a many-to-many semantic matching problem (Gu et al., 2021). Persode differs in applying personalization to memory-aware reflective conversation and illustrated diary production rather than to dialogue response quality alone. It should also be distinguished from DEPER, which addresses personalized image descriptions from attention sequences by modeling subject-specific viewing behavior and linguistic style (Xue et al., 7 Dec 2025).

Taken together, these comparisons situate Persode within a broader movement from generic interaction to persistent, user-specific systems. In Persode, that movement takes the form of a journaling environment in which onboarding, episodic memory retrieval, conversational adaptation, and visual storytelling are integrated into a single personalized reflective pipeline.