02Defining when AIproduct images aretrusted to ship
The work was not about making AI images look impressive. It was about defining the quality boundary where generated assets become safe enough for e-commerce workflows — shipped, reviewed, or rejected with clear rules.
01 — Core Story
93 test cases, 150+ AIGC analyses — mapping where AI generation is stable enough to ship
I participated in testing 93 product replacement cases and analyzed 150+ AIGC content generation cases, helping the team identify the productization boundaries of different AI paths.
Prompt + LLM
Suitable for inspiration generation, but prone to repainting and hallucination.
LoRA + LLM
More stable for same-category, similar-size product replacement.
Depth + 3D + LLM
High precision but higher cost and time — better suited as a future direction.
02 — Product Judgment
The key to shipping AI products isn't the feature list — it's the trust boundary
Which results can go live directly? Which must be manually confirmed? Which can only serve as references and never enter the distribution pipeline?
My work turned these judgments from "subjective visual inspection" into discussable, reusable evaluation dimensions: size accuracy, color fidelity, lighting naturalness, edge blending quality, and e-commerce display usability.
03 — Closing Insight
An AI PM can't just ask what the model can do — they must define under what conditions the model is worth trusting
Only when quality boundaries, human confirmation, and business workflows are clearly designed does AI capability truly become product capability.