For brand & category marketing teams

One prompt in.
A full campaign of creative out.

SOMIN reads your category off the live web, finds the territory nobody is buying, writes the propositions, produces the creative in your own Figma components — and scores every frame before a cent of media commits.

Book a brand demo Talk to sales

SOMIN · Prompt to Creative LIVE

Step 01 of 11

One prompt in

a single chat message

inputOne message naming the client, the congress and two skills. No research pack, no brand kit, no brief.

outputEverything after this is fetched by the line itself.

0/180 min machine time

Real run · MedTech congress launch, Sep 2026Client anonymized

1prompt

From brief to finished creative library

180min

Machine time, idle waits excluded

108frames

Brand-locked creatives, built in the client’s Figma

Trusted by brand & category teams across
11+ markets in APAC, MENA & Europe

P&GAudiNestléAbbottGojekKFCMothercareMcDonald’s

The creative performance equation

Campaign performance=Open territory×Brand-locked execution×Pre-spend validation

Most stacks optimise the third term after the money is committed. SOMIN moves all three before it — the territory is chosen from evidence, the execution is built from your real components, and every frame is scored while changing it is still free.

The brand-team reality

Making creative is cheap now. Knowing which creative isn’t.

Generative tools removed the cost of producing an option. They did not tell you which option to run, whether anyone in the category has already saturated it, or whether it is on brand. So volume went up, decisions did not get better, and the review meeting got longer.

The brief starts from last year

Territory chosen from the deck that worked before, not from what the category is buying media on this quarter. The open ground stays invisible.

AI output that legal won’t clear

Invented logos, drifting palettes, imagery nobody licensed. Production ends up rebuilding by hand what the model was supposed to save.

Validation happens in-market

The first honest read on a creative arrives after spend. By then the cut is expensive, and the learning lands too late for the flight it was meant to fix.

From prompt to production

Four moves, run as one line.

Each move hands the next one its inputs, so nothing is re-briefed by hand. This is the same sequence the hero walkthrough is stepping through — timings included.

Diagnose

Brand Tracker · Studies

Read the category.

Brand, palette and event scope pulled off the live web, then fifteen parallel studies across the category’s real paid output.

Palette and type read from live CSS, not a style guide PDF
15 studies, 6,502 content items, all polled to completion
Share-of-voice per territory — crowded vs. open

Decide

Insights · Propositions

Choose the ground.

Persona, tension and angle chained into a story per pillar, then propositions written only against capabilities you already own.

Whitespace verdicts, not a ranked list of adjectives
21 propositions, each traceable to a study ID
Nothing proposed that the brand cannot actually deliver

Produce

Expressions · Figma MCP

Build it on brand.

Expressions generated under a locked prompt suffix, then rebuilt as real frames in your own Figma file — optional, and only if you hand over the template link.

108 frames from your component set, not a look-alike
Licensed photography only — no invented imagery
The wordmark is composited, never drawn by a model

Validate

Scoring · Meta refresh

Judge before you spend.

Every exported frame goes back through scoring. Only creatives clearing both halves of the gate reach the live campaign — then measured CTR reopens the loop.

Predicted clickability with a confidence figure
The gate is class and confidence, never one of the two
Cut the bottom tier in review, not in-market

What one run produced

Measured on the clock, not on the calendar.

Every figure below is from a single real run. Machine time counts only the minutes the pipeline was working — waiting on a human is not a number we get to claim.

180min

Total machine time

Brand read, 15 studies, whitespace, propositions, 108 expressions, the briefing deck, Figma production and the upload — end to end.

6,502

Content items analysed

Across 15 parallel studies — two probes, six trend windows and three whitespace cohorts, all fired at once rather than in sequence.

75%

Lower briefing cost

Same on-brand quality at a fraction of the agency hours, with the strategy traceable back to the study that produced it.

Case study · MedTech × congress launch

A global surgical-technology manufacturer, one congress, one line of instruction.

The brief was a single chat message naming the client, the congress and two skills. SOMIN read the brand off the live web, ran the category, picked the territory the field was leaving open, wrote the propositions, generated 108 expressions and rebuilt them as real frames in the client’s own Figma component set.

The clinical field spends 28.8% of its paid output on safety and standards. The client was spending 3.4%. On imaging and the integrated OR, the field had bought nothing at all in 90 days — and that is where the creative went.

Global surgical-technology manufacturer · Congress launch · Paid media

Client, handles & spend anonymized — under NDA

Studies fired

15→1

All parallel, all polled to completion, one merged read.

Propositions

Written only against capabilities the client already owns.

Creatives built

108

36 per pillar — 12 upper, 12 mid, 12 lower funnel.

Figma layers

531inst.

Real component instances plus 402 text layers, in the client’s file.

Inside the whitespace read

Four territories. The two thin bars are the opportunity.

Share of the brand’s own paid output against the category’s. A territory is only worth buying when the brand is quiet and the field is loud — or when nobody is there at all.

Territory AWhitespace

Safety & standards

The field runs 28.8% of its media here. The brand runs 3.4%. The largest single gap in the read, and the one the campaign took.

Brand 3.4% vs field 28.8%NDA

Territory BEmerging

Imaging & the integrated OR

Nobody in the category bought against this in 90 days. No competition, and no proof it converts yet — a test, not a bet.

Field 0.0% in 90 daysNDA

Territory CCrowded

Training access

Half the brand’s own output already sits here against a field at 17.2%. Over-served by the brand itself — defend, don’t expand.

Brand 50.8% vs field 17.2%NDA

Territory DCrowded

Procedure literacy

45.8% of the brand’s output against a field at 2.5%. Loud, but talking to a room the category has already left.

Brand 45.8% vs field 2.5%NDA

Built for how brand teams actually work

One engine. Both sides of the room.

Whether the constraint you carry is brand integrity or media efficiency, SOMIN plugs into the point where the decision is actually made.

For brand & creative teams

On brand because it uses your components.

Production happens in your Figma file, against your real component set — so what comes out is not a model’s impression of your brand, it is your brand.

Your components, your type, your licensed photography
The wordmark composited, never generated
Palette and copy density fixed by a locked prompt suffix
Optional — the Figma step only runs if you hand over the link

For performance & media teams

Validated before the money moves.

Every frame is scored against the category corpus, and only creatives clearing both halves of the gate reach the live ad set. Then measured CTR reopens the loop.

Predicted clickability with a stated confidence figure
Both halves required — class and confidence, never one
Pushed into the aspect-matched Meta ad set automatically
Natively aligned with Meta & Google ad-diversity requirements

“Creative is the most important lever for performance in the modern advertising ecosystem. To reach diverse audiences, brands must shift from granular manual targeting to creative differentiation — letting machine learning find the right customer through the right visual message.”

— Meta for Business

Beyond human-in-the-loop

SOMIN is agentic AI-ready.

Once your workspace has accumulated brand and category signal (typically 60–90 days), graduate parts of the workflow to autonomous execution. SOMIN plugs in via MCP as the Strategy Agent in a multi-agent pipeline — which is exactly how the run in the hero was orchestrated.

Agent 01

Discovery

Discovery Agent

Reads the brand and the category off the live web — palette, scope, competitor paid output.

Agent 02

SOMIN Strategy

Strategy Agent

Runs the studies, returns the whitespace verdicts and writes the propositions behind them.

Agent 03

Visual

Visual Agent

Generates expressions under the workspace’s negative constraints, then builds the Figma frames.

Agent 04

Validator

Scoring Agent

Scores every exported frame, applies the gate, and pushes what clears it into the live campaign.

Orchestrated through Claude Agent SDK Manus Zapier Agents

Common questions

“We already have AI creative.”

Most brands do. The question is whether anything in the stack tells you which of those creatives to run. The pushbacks we hear most:

How is this different from a generative creative tool?

Those tools produce options. SOMIN decides what the option should say — the territory the category is leaving open, the proposition your brand can actually deliver, and whether the finished frame beats the category baseline. Production is one step of four, and it is the only one most tools cover.

Will it stay on brand?

Production runs in your own Figma file against your real component set, with licensed photography and a composited wordmark — nothing about the identity is generated. Palette, type and copy density are fixed by a locked prompt suffix, so drift is a configuration error rather than a creative risk.

Do we have to give you our Figma file?

No. The Figma production step is optional and gated on an explicit question — it only runs if you hand over a template link. Skip it and you still get the strategy, the propositions and the scored expressions; you just build the final frames yourself.

What does “scored” actually mean?

Each creative gets a predicted clickability class against the category corpus, with a confidence figure. Our gate requires both halves — the class must be above average and confidence must reach 0.80. A high class at low confidence does not qualify, which is the check most pre-flight scoring quietly skips.

How fast does a team adopt it?

The run in the hero started from one chat message. There is no methodology to learn and no data to prepare — name the brand, the moment and the market, and the first whitespace read comes back in minutes.

See it run on your category.

A 30-minute walkthrough on your brand, your competitors and the moment you are about to buy media against. Then — if you want it — a live session where we run the whitespace read on the competitor set you name.