The editorial team of Nordlicht Reisen GmbH (R-01) wants destination images for the catalog (R-05): a harbor in the morning, the same photo with one more sailboat, a variant in the style of an editorial hotel photo (R-04). This lesson fixes how the deployment bild is addressed and which limits are checked before the job. Paper: requirements, decision, one compiled sketch. Only the optional cloud step needs an Azure account. Cost drivers: images per generation by size and quality (for the models on the page billed as image tokens), per edit additionally the input images; the pricing page is under "Further reading". All motifs are made up and contain no people (R-12).
Prior knowledge. Filters and Prompt Shields are in "AI track, lesson T6.3", a counter around every model call in "T2.4", hard rules in code in "T6.4". Model choice is in "lesson 1.1" of this course, the SDK in "lesson 2.1", guardrails and provenance marks in "lesson 1.5". None of that is repeated; what is new is what the platform's image models can do and where their limits are.
The requirement
All numbers are assumptions of the task.
| No. | Requirement | Type | Measure | Weight | Source |
|---|---|---|---|---|---|
| 3.1-01 | Destination image from a text, landscape format for the catalog. | F | 1 image per job | Must | Editorial team |
| 3.1-02 | Change a generated image in one place, the rest stays. | F | 1 mask, rest unchanged | Must | Editorial team |
| 3.1-03 | The editorial team's hotel photo as a template for a variant. | F | up to 3 input images | Should | Editorial team |
| 3.1-04 | Every image goes through the editorial team before publication. | R | 0 images without approval | Must | Legal |
| 3.1-05 | Jobs run in the EU and through the counter (R-06, R-11). | R | location and counter per job | Must | Operations |
| 3.1-06 | Provenance marks survive processing (1.5-05). | Q | mark present after storage | Should | Editorial team |