Case study: AI product hero still

This is a composite educational case study—a realistic production path for a product hero still using the methods on this site. It is not a paid endorsement of a brand, and it does not claim a private lab test. Use it as a template for your own jobs.

Stack referenced conceptually: any strong image model + a finish tool (Photoshop/Photopea/Lightroom-class) + the creative workflow system.

The job

A small DTC kitchen brand needs a 16:9 homepage hero for a matte black electric kettle. Constraints:

Photoreal, premium-calm (not cyberpunk, not comedy)
Soft morning window light
Negative space on the right for a headline
No readable logos on the kettle (the real logo will be composited later if needed)
Must survive mobile crop of the left two-thirds

Success test: “Would we put this above the fold without apology?”

Stage 1 — Brief (written before any generate)

Line	Content
Job	Homepage hero still
Audience	Design-conscious home cooks, 28–45
Deliverable	16:9 web + ability for 1:1 social
Must include	Matte black kettle, oak counter hint, ceramic cup optional
Must exclude	Neon, busy tile, readable packaging, extra spouts, chrome fingerprints chaos

Stage 2 — References

Three real photos of matte kitchen objects with soft side light
One anti-reference: glossy stock “perfect kitchen” with harsh overheads
Palette note: warm oak, cool black, muted sage textile

Stage 3 — Generate v1

Prompt v1 (compressed for teaching):

matte black electric kettle on light oak counter,
soft morning window light from left, photoreal product editorial,
50mm shallow depth of field, muted sage cloth edge,
negative space on right, calm premium mood

Avoid: neon, logo text, busy background, extra handle, warped metal, glossy plastic toy look

Batch: 4 images.

v1 results (summary)

#	Outcome	Note
1	Kill	Neon rim light crept in
2	Maybe	Good kettle, counter too busy
3	Keep base	Clean silhouette, soft light, space on right
4	Kill	Spout geometry melted

Stage 4 — Critique (Keep / Change / Kill)

Keep: composition of #3, light direction, matte read.

Change: simplify background; slightly cooler white balance; fix micro edge aliasing.

Kill: any neon, any second product that steals focus.

Stage 5 — Refine v2

Same seed family / composition lock where the tool allowed. Prompt delta only:

“minimal uncluttered kitchen, soft north window light, single kettle hero”
stronger avoid: clutter, neon, colorful appliances

v2 batch: 3 images → one clear winner.

Optional inpaint: steam whisper above spout (subtle—steam can look fake if overdone).

Stage 6 — Finish (where it became “real”)

In a photo editor:

Crop confirmed 16:9 with right-side type safe area
Gentle contrast + warmth reduction (brand calm)
Spot cleanup on counter dust
Very light grain to reduce plastic smoothness
Export web-sized master + 1:1 center-weighted crop
Headline and CTA added in Figma, not in the model

Without finish, the frame was “nice AI.” With finish, it was a comp.

Stage 7 — Archive

Folder contents:

brief.md
prompt_v1.txt / prompt_v2.txt
seed_settings.txt
final_hero_16x9.webp
final_social_1x1.webp
postmortem.md → “Neon is banned for this brand forever.”

Next kettle job starts at Stage 2 with a stronger negative list—not from zero.

What this case teaches

Brief first prevents pretty irrelevance.
Small batches + critique beat credit burn.
One-variable refinement is how you learn the model.
Finish and real type separate demos from deliverables.
Archive is how studios get faster.

See also AI image workflows for generalized patterns and AI for design and brand when this still joins a broader board.

Variant: when the product must be exact

If the kettle SKU must match reality (handle shape, base, button placement):

Start from real product photography
Use img2img / product reference at moderate strength
Prompt primarily for scene and light
Never invent ports/buttons that mislead buyers

Accuracy jobs are composite jobs, not pure fantasy text-to-image.

Time and cost sketch (illustrative)

Phase	Time
Brief + refs	15 min
v1 batch + critique	20 min
v2 + inpaint	20 min
Finish + export	25 min
Total	~80 min

Faster than a full photo day; slower than “one prompt and pray”—and that is the point.

Checklist you can copy

[ ] Five-line brief written
[ ] Aspect locked
[ ] References gathered
[ ] v1 batch ≤4
[ ] Critique table filled
[ ] v2 one-variable changes only
[ ] Finish outside generator
[ ] Type in design tool
[ ] Archive prompt + seed + postmortem

Published by Tabaconda LLC, Florida, USA. Educational composite example—not a paid product endorsement.

QA notes from the product still

Artifacts that almost shipped: warped label edge, soft shadow direction mismatch, invented reflection of a window that wasn’t in the brief. Catch them at 100% zoom on the longest edge.

Reusable checklist

Silhouette matches client CAD/photo
No invented brand text
Shadow consistent with key light
Background seamless for ecommerce crop

Published by Tabaconda LLC, Florida, USA.

Useful physical tools for creator tools work. Optional product searches (see disclosure). Buy only what you need.

🎨 Back to Studio

Case study: AI product hero still