Your brand DNA, in every AI-generated look
PromptiQ learns your brand from the catalog you already own, then scores every AI-generated image against it on a calibrated Brand Fidelity Score. On-brand becomes a number, not an opinion — and a self-improving loop keeps every look locked to your house style.
Built with early fashion design partners — grounded in real embedding science, not prompt guesswork.
a calibrated score on every image — product + brand centroid
jina-clip-v2 visual embeddings behind the metric
each approved look banks into your Brand Memory
train your own brand model, quality-gated before it ships
From catalog to on-brand output in three steps
Not another prompt box. PromptiQ learns your brand from the imagery you already own, then makes that DNA reusable.
Upload your catalog → Brand DNA
Drop in the product photography you already have — a single drop or your whole archive. PromptiQ reads every image with a fashion-trained vision model and distills a structured Brand DNA profile: visual style, palette, voice, signature elements, photography style, and an avoid-list that steers every generation.
Optimize with the self-improving loop
Each image becomes a draft prompt, gets regenerated, and is scored on the calibrated Brand Fidelity Score — measured against both the target product and your catalog-wide brand centroid. A creative-director critique feeds a textual-gradient refinement, loop after loop, converging on target fidelity in fewer iterations and lower cost per approved image.
Generate on-brand — and bank it
Every approved, high-scoring result is banked into your Brand Memory as an exemplar, so future generations retrieve your own proven prompts and vocabulary. Generate fresh looks at fast, balanced, or quality presets — and the product gets more on-brand the more you use it.
Fluid ivory linen midi dress, relaxed drape, natural side light, editorial resort styling…
used 12×on-brandOn-brand is a number, not an opinion
Every image is embedded with jina-clip-v2 into a 512-dimensional visual space, then scored against both the target product and your catalog-wide brand centroid — a 0.7 / 0.3 blend. The result is calibrated so the score actually tracks perceived on-brandness: genuinely off-brand output floors out, strong on-brand work lands in the 85–100 band. A measurement, not a vibe.
- jina-clip-v2, 512-dim, scored per image
- Product match (0.7) + brand centroid (0.3), calibrated
- Genuinely off-brand output scores low — no false confidence
It gets more on-brand the more you use it
Every approved, high-scoring result is banked as a brand exemplar — its winning prompt, its embedding, its fidelity. On every later generation, PromptiQ retrieves the exemplars most visually similar to the new product and reuses your own proven prompts and vocabulary. The product compounds into a per-customer data moat that competitors can't copy.
- Approved looks bank as reusable brand exemplars
- Future generations retrieve your own proven prompts
- A per-customer moat — your brand's accumulated memory
Ivory linen resort dress
Tailored wool overcoat, editorial
Silk slip, soft studio light
Optimize an entire catalog — and watch it happen
Kick off a batch and PromptiQ works through every asset, streaming live progress: which image is running, its current iteration, and its climbing Brand Fidelity Score, all under one overall completion bar. Walk away and it emails you when the run is done.
- Batch optimize a full catalog in one run
- Real-time per-asset iteration + fidelity updates
- Completion email with average fidelity and duration
Train your own brand model
On the premium tier, your catalog fine-tunes dedicated diffusion weights (LoRA) on GPU infrastructure — a model that generates in your house style natively. A self-verifying quality gate only promotes it if it beats the prompt-only baseline on the very same Brand Fidelity metric (a live run: 90.6 vs 86.6). Then it's one click in the app: toggle it on and compare side by side.
- Your catalog trains dedicated LoRA weights on GPU
- Promoted only if it beats the prompt-only baseline
- One-click toggle + side-by-side compare in the app
prompt-onlyyour modelStop shipping generic. Start shipping on-brand.
Left to a raw model, AI imagery drifts into the same glossy sameness everyone else ships. PromptiQ scores every result against both the target product and your catalog-wide brand centroid on a calibrated Brand Fidelity Score — so “on-brand” stops being a feeling and becomes a number you can hold the output to.
Off-palette, off-silhouette, off-brand
On-brand band: 85–100 · calibrated, not a raw cosine
Consistent house style, measured every time
On-brand band: 85–100 · calibrated, not a raw cosine
the calibrated Brand Fidelity band on-brand work lands in — raw drift scores far lower
product-match and brand-centroid blend behind every fidelity score
a Brand Studio model must beat the prompt-only baseline before it's ever served
Bring your catalog. Keep your house style.
PromptiQ is onboarding a small group of fashion teams as early design partners. Get set up with your own Brand DNA, calibrated Brand Fidelity scoring, and — on the premium Brand Studio tier — a brand model trained on your own catalog.
No card required. We’ll help you optimize your first catalog personally.
What early partners get
- Hands-on onboarding with your first catalog
- Your own brand-DNA library + calibrated Brand Fidelity scoring
- Early access to Brand Studio — train your own brand model
- Direct line to the team building it
Frequently asked
Keep your brand out of the AI sameness
Turn your catalog into a Brand DNA the AI can’t drift from, and score every look on a calibrated Brand Fidelity Score — with a flywheel that gets more on-brand the more you use it.