Food Companion

After logging one meal, what makes the next entry worthwhile?

An existing landscape, with different experience priorities

I compared public product descriptions with the archived research to examine logging, interpretation, and relationship participation. These are qualitative design judgements, not rankings or claims that unreviewed features are absent.

Source check: 11 October 2026. BiteSnap and Kaimai retain archived research observations; their latest versions were not tested and their public pages were not reliably accessible in this check. Food Companion’s proposed distinction is explainable feedback connecting personal records with optional care between familiar people.

BEHAVIOUR OBSERVATION / CHANNELS

How content becomes care

Xiaohongshu: Observe how health content is understood, saved, and forwarded as a conversation starter.

Douyin: Observe participants’ existing sharing and response habits. Shared objects such as streak characters can connect people while creating pressure to keep a streak.

Channel selection follows the research questions and participants’ contexts; no quantitative channel comparison was conducted.

Enlarge image

1. Share a meal

   Use everyday life as a low-effort starting point

2. Receive a response

   Make the logging effort feel acknowledged

3. Keep a shared object

   Let pets and space hold a visible history

4. Preserve choice

   Make invitation, skipping, and pausing natural

Three breaks lead to four principles: an easier start, clearer feedback, a non-judgemental response, and choice in relationship participation. Personal logging is complete in itself; companionship extends it.

The complete loop around one entry

1. 01Meal input
2. 02Confirmation
3. 03Pet response
4. 04Record explanation
5. 05Diary review
6. 06Optional sharing
7. 07Companionship & home accumulation

A journey grounded in everyday moments

01

A busy lunch

Start one entry

02

Capture / import

Use a photo instead of repeated input

03

Review and save

Correct title, meal type, and portion

04

Understand the response

Inspect the confirmed-record basis

05

Diary and collection

Reflect and consider the next meal

06

Optional sharing

Invite someone and build a shared history

The journey describes research-informed scenarios and intended experiences, rather than measured satisfaction or an emotional score.

Less effort to log; the final confirmation stays with the user.

01Start with one meal02Review candidates03Adjust the portion04Save after confirmation05Return to the diary06Collect the meal

Enlarge image

01 / 06

Start with one meal

Take or import a photo, with photo-only logging available. The camera is simulated; file import is implemented in the local prototype.

Original Figma UI · Enlarge to read the complete screen

Recognition proposes; confirmation records

A photo reduces input but cannot guarantee correct food or portions. Review comes before saving, with correction available and photo-plus-text logging retained when AI is unavailable.

This connects completing an entry with receiving an understandable response: diary history, collectible stickers, and an optional next-meal plan.

Existing style assets, not results of a live recognition or sticker-generation test.

The pet first acknowledges individual participation, then extends care to a willing companion. Each person has their own pet and records. A shared space connects people without confirming entries for them or merging body data.

Enlarge image

Enlarge image

Enlarge image

Four expressive directions need an understandable basis

Energy

Activity and rest in the character, not inferred human fatigue.

Satiety

A post-meal expression; actual fullness requires self-report.

Digestion

A proposed post-meal rhythm; mapping rules remain unvalidated.

Sleep

An evening rest cue, not an assessment of sleep quality.

These pair expressive directions with existing artwork; they are not four validated physiological mappings.

Enlarge image

Explain the response

For example, a confirmed evening drink and time can prompt a designed rest cue, with its record basis and optional next step. A sleepy-looking character does not mean the system has diagnosed insomnia.

Mood is expressed by the user. Missing entries are not automatically interpreted as sadness, anxiety, or poor health.

Enlarge image

Enlarge image

Enlarge image

Let continued participation leave a place behind.

01Make participation visible02Claim a reward03Browse the catalogue04Arrange and resize05Save and return home

Enlarge image

01 / 05

Make participation visible

Daily participation leads to furniture accumulation, with unlock conditions in the catalogue. Simulated days test the reward path, rather than long-term retention.

Original Figma UI · Enlarge to read the complete screen

32 furniture assets, one soft visual language

Enlarge image

Enlarge image

Enlarge image

Enlarge image

Enlarge image

Enlarge image

Enlarge image

Enlarge image

Eight selected assets from the 32-item catalogue, checked against the engineering map. Some Figma category labels still differ from the runtime prototype.

Design judgement runs through the process; AI helps turn decisions into inspectable outputs.

This chapter explains how I used AI in asset production and implementation. Food recognition, nutrition processing, and sticker generation inside the app are a separate product layer whose live results still require validation.

1. 01 / Human-ledDefine direction

   Problem, research, rules, architecture, visual constraints

2. 02 / AI assistance + human judgementProduce and refine

   Prompt → generate → select and correct → consistent assets

3. 03 / Figma → Codex → GodotImplement interaction

   Screens → scenes / components → scripts → local prototype

4. 04 / Human review + automated checksCheck and iterate

   Runtime comparison → diagnose → revise → regression check

01 / DESIGN DIRECTION

Define what matters, then what counts as a good result

I defined confirmation-before-recording, explainable responses, and optional sharing, then organised the record, interpretation, and relationship layers. AI did not replace research conclusions, hierarchy, or final visual judgement.

Cream and muted green unify flat scenario illustrations, the soft pet, and furniture. A large head, short body, cream muzzle, dark nose and paws, peach cheeks, and two green leaves maintain recognition.

Design input

Research breaks and relationship boundaries

Inspectable output

Flows, UI states, components, and asset constraints

Enlarge image

02 / AI-ASSISTED ASSET PRODUCTION

From generated output to consistent product assets

The archive retains illustration prompts, initial outputs, and eye-style revisions. I set constraints, selected usable results, and checked that local changes preserved composition, colour, and character recognition.

Enlarge image

Enlarge image

ACTUAL PROMPT RECORD · EXCERPT

“muted forest green #527B64, sage #CFDCC6, mist blue #DDEAF0, warm cream and peach skin”

Revision constraint: solid dark-green vertical oval eyes; no sclera, highlights, or eyelashes. From assets_v8/prompts_final.json.

1. Constrain

   Proportions, style, colour, alpha edges

2. Generate

   Specify the scene and purpose

3. Select and revise

   Preserve composition; fix inconsistencies

4. Organise and integrate

   Name, check dimensions, reuse in UI

03 / FIGMA → CODEX → GODOT

From visual states to running interaction

I defined screens and states in Figma, then used Codex to help build Godot scenes, reusable components, and scripts. The central question shifted from visual similarity to consistent input, confirmation, persistence, and recovery.

Enlarge image

Enlarge image

ORIGINAL SOURCE · food_flow.gd

Update the same entry without double-counting

	var updated := records.filter(func(item): return item.id != id)
	updated.push_front(saved)
	if not write_records(updated): return false
	records = updated
	last_saved = saved.duplicate(true)
	view.record = saved
	app.show_page("165:1510" if planned else "165:1308")
	return true

Original excerpt. Filter and update by existing id; update memory and navigate only after writing succeeds.

Pages & components

main.gd / surface.gd

Routing, history, overlays, native Control layout

Meal data chain

food_view.gd / food_flow.gd

File import, confirmation, persistence, summaries

Spatial editing

room_editor.gd / room_canvas.gd

Dragging, scaling, layers, drafts, recovery

The archive supplies Figma exports, scene and script files, and runtime captures. It does not include a separate editor-operation capture; no simulated editor image is substituted.

The project moves from product rules, screens, and assets into a local interactive prototype. Implementation makes design decisions operable and inspectable; the number of tools used is not the outcome.

Enlarge image

Enlarge image

Enlarge image

Next design iteration: from a meal response to a shared plan

Enlarge image

Enlarge image

Enlarge image

These Health 3.0 designs are from the latest Figma export and are not yet in the Godot prototype. Original app artwork retains Chinese; this case’s explanations, controls, and navigation are bilingual.

Make an entry feel acknowledged, while keeping the user in control.

This case uses actual screens, assets, and runtime captures from the latest archive. Anonymous access to the Figma prototype and public repository was verified on 11 October 2026. Godot remains a local prototype with no Web release.