🎯 Blind Comparison Scheme
Prepare 20 real digital-printing scenario tasks: DTF prints / UV stickers / sublimation / T-shirt mockups. The same prompt runs on the local solution and on Lingtu; both result sets are shuffled and numbered, then scored blind by staff or the boss who do not know the source. All scoring sheets are archived.
Composition of the 20 tasks：DTF prints × 5 · UV stickers × 5 · Sublimation × 5 · T-shirt mockups × 5
Scoring Dimensions
| Dimension | Weight | How to score |
|---|---|---|
| Fidelity | 40% | How closely pattern detail, color, and style match the prompt target (out of 100). |
| Usability | 30% | Whether it enters the prepress pipeline directly: resolution / background / bleed pass without rework. |
| Speed | 15% | Time per image (local solution including GPU queue time). |
| Cost | 15% | Cost per image: local ≈¥0 electricity vs Lingtu subscription / pay-per-image. |
✅ Pass Threshold Definition
- Combined score ≥ 90% of Lingtu's score = judged 'equal and usable';
- The fidelity dimension must not fall below 80 (out of 100);
- At least 16 of the 20 images pass usability.
If it fails, go back to P2 for workflow tuning — it is never forced live. Until acceptance passes, this solution is an experiment only and replaces no existing process.
PENDING · awaiting boss approval
⚠ The boss is asked again before every single step starts