System intelligence · live engine
See how the engine actually works.
Every extraction runs through a real, tiered analysis pipeline. This page shows the capability profile of each plan, the live model ecosystem behind it, and the exact stages the engine walks through - no gloss, no mock data.
Capability profile
Three plans. Three depths.
Each tier runs the same engine at a different depth. The radar below plots the real capability profile of each plan - how much the pipeline extracts, how precisely, and how fast.
Interactive
Nano Lite
Fast, concise extraction - the everyday mode. The engine reads the subject, scene, lighting, colors and mood, then writes a clean working prompt of up to 1,000 characters.
Model selection
Your key. Your model. One engine.
We don't own models - and we don't pretend to. The engine runs on vision LLMs you connect with your own OpenRouter key and routes every extraction to the model you select. Free models already deliver advanced reverse engineering; paid models take it to full precision.
Selectable through your key
Top selectable categories by model count - all connected through your own key. Full list on the model explorer.
The pipeline
What the engine does on every image.
Each tier walks a real, fixed stage list. The deeper the tier, the more the engine decomposes before it writes a single word of the prompt.
Nano Lite · Standard
6 stagesImage enters the pipeline, decoded and normalized.
What is in the frame, who is there, where they are.
Subject, setting, lighting, colors, mood - the essentials.
One clean, copy-ready prompt up to 1,000 characters.
This is the free tier. It runs on every plan.
Nano Spark · Deep
10 stagesImage enters the pipeline, decoded and normalized.
Scene, subjects, environment and relationship.
Framing, placement, hierarchy, negative space.
Direction, softness, shadows, temperature.
Pose, expression, textures, surfaces, products.
Layout, typography and hierarchy when visible.
Structured prompt up to 2,000 characters.
Consistency check before delivery.
Professional reverse engineering - Pro and Ultra plans.
Spark Ultra · Ultra
15 stagesImage enters the pipeline, decoded and normalized.
Full reverse-engineering pass over the frame.
Every layer, region and distinct visual area.
Framing, lens character, perspective, depth.
Every light, shadow and surface response.
Facial detail, hands, clothing, strand-level.
Typography, frames, logos - verbatim text.
Position inventory - every element placed.
Layer-by-layer reconstruction spec.
Exhaustive prompt up to 20,000 characters.
Quality-control pass re-checking the image.
Corrections folded in, output verified.
Deepest reconstruction - Spark Ultra plan.
Live depth, live limits
per planNano Lite limits - server enforced.
Compare all plans →The progression
From extraction to exact reconstruction.
Extract & understand. Quick, practical extraction and editing - a working prompt in seconds.
Reverse engineer & restructure. Professional depth - layout, typography, hierarchy, deeper lighting and materials.
Deeply transform & rebuild. Layer-by-layer decomposition with a quality-check pass, up to 20,000 characters.
Everything from Ultra plus video reverse engineering when it launches.
Now see it on a real image.
Install the extension, sign in, and hover any image - the engine runs the exact pipeline above, live.