How We Test and Score
Every ranking on Verdict Lab comes from the same battery. No exceptions, no exceptions for advertisers (we have none) and no exceptions for friends.
Dimensions and weights
| Dimension | Weight | What it measures |
|---|---|---|
| Output quality | 40% | Blind-scored realism, composition and artifact rate across 50 standardized prompts |
| Prompt adherence | 25% | Share of prompts where all constraints (count, text, spatial relations) render correctly |
| Editing control | 20% | Inpainting stability, style consistency across a series, iteration workflow |
| Value for money | 15% | Per-image cost at production volume vs quality delivered (published pricing, verified quarterly) |
Test protocol
Each tool receives the same 50 prompts (available on request), run at default settings in October 2026. Three reviewers score outputs blind — no tool names attached. The final score is the weighted mean. You can recompute any score on this site from these rules.
Independence and disclosure
We buy our own test subscriptions. Manufacturers do not see scores before publication. Verdict Lab is part of a network of independent review publications — see the about page for the full list of related publications, which do not represent independent confirmations of one another.