104
Product images
From five supplier photos
AI content creation · Case study
AI Product Image System
Filtered shower head · brand withheld19 August → 8 September 2026
104
Product images from five supplier photos
41 catalogue · 63 campaign
01
Context
A water-filtration brand needed imagery for its website, social media and future photo/video generation.
The only input was five 487×508 px supplier photos: four colourways plus a cartridge, with spec text burned into the images.
Role in the source: AI content creator, prompt engineer and visual producer.
02
Constraint
The models had never seen this exact product. A beautiful image was useless if the shower geometry changed.
- Image models had never seen this exact shower geometry: an oval ring with a central hole, two different faces and water that exits inward.
- The source rule was “fidelity before style”: invent nothing, swap nothing, geometry is law.
03
Strategy
01Forensic tech card
02Angle catalogue
03Reference sheets
04Product lock + negatives
05Scene, light + water
06Fidelity check
04
Execution
The work started with product fidelity: measure the object, lock the geometry, separate the two faces, then vary angle, light, water and scene.
- A G1/2-inch thread (≈21 mm) became the ruler for a 50+ parameter tech card; sizes are approximate because they were measured from photos.
- The angle catalogue expanded one source view into ten per colour; the catalogue total was 41 frames including the cartridge.
- Reference sheets separated the two product faces and removed burned-in text that caused pseudo-glyphs.
- A product lock fixed the oval geometry, two rows of micro-holes, handle proportions, thread and face rules across prompts.
- A 13-preset light library and dedicated water module controlled scene variation without changing product geometry.
- Every prompt shipped with a fidelity report; if “Invented” could not read NONE, the prompt was fixed.
05
Results
41
Catalogue frames
Angles and faces across colourways
63
Campaign frames
Range, hands, editorial, studio, water and off-brief tests
17
Reference sheets
One job each; text kept out of model inputs
The system included 13 lighting presets, 7 working documents and prompt registers for 13 image models.
The 63 campaign frames were generated across seven generation days from 19 August to 8 September 2026.
The content strategy mapped imagery to reach, trust, desire and decision.
The reusable kit included the tech card, photo catalogue, prompt kit, water module, multi-model studio architect, light library and brand-colour reference.
06
Evidence
Selected outputs
Campaign image system
Selected generated frames showing product colourways, editorial portraits, water scenes and controlled campaign compositions.
07
Proof limits
What this case does not claim.
- No sales, reach or conversion figure is attached to this production work, so none is claimed.
- Sizes were measured from photos and are approximate.
- Every face in the set is generated.
- 4 of 63 campaign frames were off-brief; the brand name is withheld.
08
How we worked
The deliverable was a reusable kit, not a one-off image batch: tech card, catalogue, campaign set, prompt system and handover.
- Client sends product photos and brand rules if available.
- Delivery includes a tech card and catalogue, campaign set and prompt kit.
- One handover and one review per batch are the client-time model.
- The kit is designed to stay with the team after the set is done.
The final system is designed for products that have not been shot yet and to support future product-photo and video generations.
View full presentation
1 / 39
Swipe on mobile or use the controls.
Next step
Work with me on UpWork
If this case is close to your problem, send me the context on UpWork.
View my UpWork profile