Complete User Guide

From Catalog to Brand-Aligned Images

This guide walks you through the full PromptiQ pipeline — uploading your product catalog, running AI prompt optimization, and generating new designs that stay true to your brand identity.

Contents

  1. Pipeline Overview
  2. Step 1 — Create a Catalog
  3. Step 2 — Upload Product Images
  4. Step 3 — Run Prompt Optimization
  5. Step 4 — Generate Images from Optimized Prompts
  6. Step 5 — Manage Concepts
  7. Step 6 — Organize with Collections
  8. Step 7 — Prompt Library
  9. Step 8 — Sharing
  10. Settings & Customization
  11. FAQ

Pipeline Overview

PromptiQ uses a two-stage AI pipeline. First, it learns what makes your brand visually distinctive by analysing your existing product images. Second, it uses that knowledge to generate new images that look and feel on-brand — without you having to write a single prompt by hand.

📂

Stage 1 — Prompt Optimization

Upload your catalog images. AI analyzes each one and iteratively refines a text prompt until the re-generated image closely matches the original (measured by CLIP similarity score).

🎨

Stage 2 — Image Generation

Use the optimized prompts to generate new brand-aligned images on demand. Choose Fast, Balanced, or Quality presets to control speed vs. fidelity.

Upload Images→CLIP Embeddings→AI Optimization Loop→Optimized Prompts→Generate Images→Concepts Gallery
1

Create a Catalog

A catalog groups your brand images together. You might create one catalog per collection, season, or product line.

  1. 1.Go to Catalogs in the left sidebar and click "New Catalog".
  2. 2.Give it a name (e.g. "Spring Collection 2026") and an optional description.
  3. 3.Click "Create" — you will land on the empty catalog page.
💡 Tip:Keep catalogs focused — 10 to 50 images per catalog gives the optimizer enough variety without diluting the brand signal.
2

Upload Product Images

Each image becomes an asset — a source of visual brand information that the optimizer will learn from.

  1. 1.Open your catalog and click "Upload Assets".
  2. 2.Drag and drop JPG or PNG files, or click to browse. Up to 100 images per catalog.
  3. 3.Images are uploaded to Supabase Storage and a CLIP embedding is extracted automatically.
  4. 4.Once uploaded, each asset appears in the catalog grid with its filename.
ℹ️ Note:Images are analyzed using CLIP (Contrastive Language-Image Pretraining) to create a numerical fingerprint of each photo's visual content. This fingerprint is what the optimizer tries to match when generating prompts.
💡 Tip:Use clean, high-quality product shots rather than lifestyle images for the most accurate optimization. Plain or neutral backgrounds work best.
3

Run Prompt Optimization

This is the core of PromptiQ. The optimizer runs an iterative AI loop for each asset until it produces a prompt that can recreate your image as accurately as possible.

How to start

  1. 1.Open your catalog and click "Optimize All" (or select specific assets and click "Optimize Selected").
  2. 2.Choose the number of iterations (3–10). More iterations = more refined prompts but longer run time. 5 is a good default.
  3. 3.Set a target similarity score (0.7–1.0). The optimizer stops early if this threshold is reached.
  4. 4.Click "Start Optimization". You will be taken to a live progress page.

What happens under the hood

Initial prompt

Gemini analyzes the original image and drafts a descriptive starting prompt.

Generate

Together AI (Seedream-3.0 or HiDream-I1) generates an image from the current prompt.

Score

Jina CLIP compares the generated image to the original using cosine similarity (0–100%).

Critique & Refine

Gemini reviews the similarity score and the current prompt, then rewrites it to close the gap.

Repeat

Steps 2–4 repeat until the target similarity is reached or max iterations are exhausted.

ℹ️ Note:A typical run of 10 assets at 5 iterations takes 5–15 minutes. The progress page polls automatically — you do not need to keep it open, but it will show live updates if you do.
💡 Tip:Similarity scores of 70–85% are excellent for fashion photography. Scores above 90% are rare and indicate the prompt almost perfectly captures the original.
4

Generate Images from Optimized Prompts

Once optimization is complete, your prompts are ready to generate new brand-aligned images on demand.

From Catalog (recommended)

  1. 1.Go to Generate in the sidebar and click the "From Catalog" tab.
  2. 2.Select your catalog from the dropdown — all optimized prompts appear, each labeled with the source asset filename.
  3. 3.Click any prompt to pre-fill the input field. You can edit it freely before generating.
  4. 4.Choose a quality preset (see below) and click Generate.
  5. 5.The new image appears instantly and is saved to your Concepts gallery.

Manual Prompt (freestyle)

Use the "Manual Prompt" tab to write your own prompt from scratch. Useful for exploring concepts outside your catalog.

Quality Presets

FastSeedream-3.0

Best for rapid iteration and exploring ideas. Generates in ~10 seconds.

BalancedGemini 2.5 Flash

Google Gemini as primary. Excellent for fashion — strong understanding of garment detail.

QualityHiDream-I1-Full

Highest fidelity output. Best for final hero shots. Generates in ~60 seconds.

5

Manage Concepts

Every generated image is saved as a concept. The Concepts gallery is your visual working space.

Filter & Search

Use the "All Catalogs" dropdown to see only concepts linked to a specific catalog. Toggle "Favorites only" to focus on your starred images.

Lightbox View

Click any concept to open a full-size lightbox showing the image, prompt used, model, quality preset, and dimensions.

Notes

Add, edit, or delete notes on any concept directly from the lightbox. Useful for feedback, revisions, and creative direction.

Favorites

Star any concept to add it to your favorites. Favorited concepts are highlighted and can be filtered independently.

6

Organize with Collections

Collections let you group concepts into curated sets — a mood board, a client presentation, or a shortlist for sign-off.

  1. 1.Go to Collections and click "New Collection". Give it a name and optional description.
  2. 2.In the Concepts gallery, select multiple concepts using the checkboxes that appear on hover.
  3. 3.Click "Add to Collection" and choose your collection.
  4. 4.Open the collection to see all concepts in a grid. Remove individual concepts or delete the entire collection when done.
7

Prompt Library

Save prompts you want to reuse or share with your team. The Prompt Library is separate from catalog-linked prompts — it is a personal or team-wide collection of your best creative starting points.

Save a Prompt

Go to Prompts → "New Prompt". Give it a title, paste the prompt text, add a category (e.g. Fashion) and comma-separated tags.

Search

Use the search bar to filter by title, prompt text, or tag. Results update instantly.

Use a Prompt

Click "Generate" next to any saved prompt to open it pre-filled in the Generate page. The usage counter increments so you can see which prompts are most popular.

Usage Tracking

Each prompt shows a usage count. High-count prompts are your team's favourites and worth keeping polished.

8

Sharing

Share individual concepts with stakeholders — no login required for the recipient.

  1. 1.Open a concept in the lightbox and click the "Share" button.
  2. 2.A unique URL is generated (e.g. https://www.promptiq.cloud/share/abc123).
  3. 3.Copy and send that link to anyone. They will see the image, prompt, model, and dimensions — no sign-in required.
ℹ️ Note:Share links display a "Shared View (Read-Only)" banner so viewers know they cannot edit the concept. Each view increments a view counter on the share record.

Settings & Customization

Access Settings from the sidebar to customize your workspace.

General tab

  • •Theme — Light, Dark, or System
  • •Default Target Similarity — the score the optimizer aims for (0.7–1.0)
  • •Default Max Iterations — how many refinement loops to run (3–10)
  • •Changes are saved to your browser and applied to all future optimization jobs

Team tab

  • •Invite team members by email with a Designer or Viewer role
  • •Manage pending invitations
  • •All members share the same organization workspace and catalog data

API Usage tab

  • •See total images generated, tokens used, and estimated cost
  • •Breakdown by provider (Together AI, Google) and model
  • •Daily usage chart for the current period

FAQ

How long does optimization take?

A single asset at 5 iterations takes roughly 1–3 minutes. A full catalog of 10 assets runs in parallel and typically completes in 5–15 minutes. The progress page updates every 30 seconds.

What similarity score should I aim for?

70–85% is excellent for fashion photography, where lighting, background, and pose variation are expected. Scores below 60% usually mean the prompt needs more specific garment description. Scores above 85% are very high-fidelity.

Can I edit an optimized prompt before generating?

Yes. In the Generate → From Catalog tab, clicking a prompt pre-fills the input field where you can freely edit it before hitting Generate.

What image formats are supported for upload?

JPG and PNG. For best results use product images at least 512 × 512 px. The CLIP embedding model works best on clean, well-lit product photography.

Are generated images stored?

Yes — every generated image is automatically saved as a concept in your Concepts gallery, linked to the prompt and model used.

Can I re-run optimization on assets that already have prompts?

Yes. Select the asset in the catalog and click Optimize. The optimizer will start fresh and overwrite the previous prompt if it achieves a higher similarity score.

How do I share the whole catalog, not just one image?

Individual concept sharing is currently supported. For catalog-level sharing, use Collections — group the generated concepts and share individual ones from within the collection.

Ready to build your prompt catalog?

Sign in to your workspace and start with your first catalog upload.

Go to DashboardCreate Account