AI Gift Generation Production System
A structured AIGC production pipeline for themed gift concept generation — from creative brief and prompt components to generation, tagging, scoring, review, and reusable asset storage.
01 · Overview
What the system is
AI Gift Generation is a self-initiated production system, not a single deliverable. It turns ad-hoc requests for themed gift concepts into a structured, repeatable pipeline that can be run, evaluated, and scaled.
I designed the workflow around generation: how creative requests become structured briefs, how briefs become reusable prompt components, how raw outputs get tagged, scored, reviewed, and finally stored as reusable production assets — with a feedback loop that improves the system over time.
Context
Themed gift concepts are a fast, high-volume creative surface: campaigns constantly need fresh, on-brand concepts tied to themes, events, and audience signals.
Generative AI made individual images cheap, but production needed structure — consistency across many outputs, traceability of what changed, objective review, and a way to reuse what already worked.
02 · Problem
An ad-hoc creative process that couldn't scale
Old workflow
- 1Creative request
- 2Manual ideation
- 3Reference search
- 4Prompt trial
- 5Batch generation
- 6Manual review
- 7Feedback
- 8Asset delivery
Why it broke down
- Inconsistent quality across outputs
- Weak traceability — hard to tell what changed
- Subjective, slow review
- Hard to reuse past results
- Hard to scale beyond a few requests
- Bad cases were hard to diagnose
03 · System
A structured AIGC production system
The new design connects every step from brief to feedback loop. Select any node to see its input, the transformation it performs, its output, and why it matters.
Structured brief
Input
Ad-hoc creative request from a campaign or theme.
Transformation
Capture intent, scenario, and constraints in a structured template.
Output
A reviewable, comparable brief.
Why it matters
Structure makes every downstream step repeatable.
04 · Prompt & Evaluation
Prompt system and evaluation, by design
Prompt design here is a system, not one clever sentence. Each request is assembled from components that separate creative intent, constraints, and generation parameters — so prompts are testable and reusable.
- 01Subject
Core concept of the gift
e.g. Festival lantern gift, celebratory
- 02Audience signal
Persona / theme / audience cues
e.g. Cozy lifestyle theme, warm tone
- 03Visual style
Rendering and art direction
e.g. 3D, soft studio light, glossy
- 04Object / symbol
Key motif to feature
e.g. Gift box + ribbon + sparkle
- 05Color logic
Palette rules tied to theme
e.g. Warm reds + gold accents
- 06Platform constraints
Aspect, safe area, legibility
e.g. Square, centered, mobile-safe
- 07Negative prompt
What to avoid
e.g. No text, no clutter, no logos
- 08Generation params
Model / steps / seed control
e.g. Fixed seed for comparison
- 09Review criteria
How it will be judged
e.g. On-theme, clean, production-ready
Evaluation & review
Evaluation is designed into the workflow, not bolted on at the end. A weighted scoring model plus a clear review status make approval objective and bad cases diagnosable.
- 01Aesthetic quality
Composition, lighting, and finish at production standard.
- 02Consistency
Coherent with the theme and with sibling outputs in the batch.
- 03Visual relevance
Matches the brief, audience signal, and intended symbol.
- 04Safety
No policy, brand, or content risks.
- 05Production readiness
Usable as-is: aspect, safe area, legibility on mobile.
Bad cases are diagnosed against the same dimensions, then fed back into prompts and criteria — closing the iteration loop.
05 · Asset Reuse
From raw outputs to a reusable library
Raw AI outputs only create value if they can be found and reused. Every approved asset carries the metadata needed to search, filter, and reuse it.
06 · Impact
What it produced
My role
- Translated ambiguous creative requests into structured design inputs.
- Designed prompt components that separate creative intent, constraints, and generation parameters.
- Connected generation outputs to tags, scoring criteria, review status, and reusable asset storage.
- Designed evaluation as part of the workflow, not as a final subjective check.
- Defined the benchmark / QA review approach and diagnosed bad cases.
- Kept the creative, data, and generation logic coherent as one system.
This is a self-initiated concept — outcomes are qualitative and illustrative, not company metrics.
07 · Reflection
Designing the workflow, not the visuals
AI workflow design is different from visual design: the deliverable is the system that produces visuals reliably, not any single image.
Evaluation has to be designed early. If you can't score and review output objectively, you can't scale generation or diagnose failures.
A prompt system becomes a production system when its components are reusable, traceable, and tied to evaluation and storage.
Reusable assets compound — they lower the cost and raise the consistency of every future request.
Want to see the prototype logic behind this system?