Methodology · Updated: 08 September 2026

How we review undress AI tools

Five weighted criteria, dated test notes and a consent rule — the method behind every score on this site, in English.

Weights

Five criteria and their weights

The overall score is a weighted average on a 0–10 scale. The weights are fixed for every tool, published here and on the Polish page Jak oceniamy, and have not changed since the first test.

CriterionWeightWhat we measure
Privacy30%retention, deletion, payment data
Quality25%consistency, artefacts, HD output
Price20%cost per image, free tier
Speed15%generation time, queue
Support10%help desk, documentation, refunds

Weights sum to 100%. A tool with the best image quality can rank below a cheaper one that handles data better — this is intended.

Privacy (30%). We read the privacy policy and terms before paying, then check four points: how long uploads are kept, whether deletion can be requested and is confirmed, which payment data the operator collects (card, crypto, invoice details), and whether the terms are readable before checkout.

Declared values are marked as declarations — “deletion within 48 hours (operator’s declaration)” — because we cannot audit servers.

Quality (25%). Ten control images from the team’s own material or synthetic sources, identical for every tool: front-facing, side pose, loose clothing, complex background, low light. We rate consistency across the set, artefacts on hands and faces, native resolution and whether HD sits behind a paywall or a watermark.

Price (20%). We convert every plan to an effective cost per image in the operator’s currency on the test date: credits per image, pack price, subscription allowance. Free tiers count only for what they actually deliver — one watermarked image per day is a trial, not a free plan. Prices hidden until checkout cost points.

Speed (15%). Median generation time over the control set, measured with a paid account and — where one exists — a free account, because queue policy is the main difference. Support (10%). Response time to one written question, documentation quality, refund terms and whether a legal entity and contact address are published.

How a test runs, from screening to publication

A tool enters the catalogue after legal screening: we check that the operator publishes terms, blocks content involving minors and offers a way to report abuse. Tools that fail screening are not reviewed at all; tools that pass but have not been tested are listed as “test scheduled” with no score, no estimated price and no outbound button — currently Undress.zone, Yollo.ai, Pixlr and Unclothy.

The test itself takes one working day per tool. We pay for the entry plan, run the control set, time each generation, ask support one question and read the policies again after payment. Everything is written into the test notes with the date; the published review quotes those notes rather than marketing pages. The current scores were recorded between 12 and 29 August 2026.

Each review is then checked against the site’s rules: no invented statistics, no user reviews or testimonials, no named author profiles, and a consent note next to every rating. The Polish editorial policy spells these rules out; the English section follows the same document.

Scores are not averages across the network. Where our earlier tests on sister editions produced a different number for the same tool, we keep the value with the full per-criterion breakdown and note the discrepancy in the review rather than blending them.

Independence, affiliate links and what we are not

Six of the sixteen tools in the catalogue — Ainudez, PornWorks, Deep-Undress, UndressWith, WaveSpeed and UndressHer — are reached through affiliate links. Every such link is routed through /go/, carries the sponsored attribute and sits next to a visible disclosure. Ten tools have no commercial relationship; their outbound link is a plain redirect and is labelled “not an affiliate link”.

Three of them — Dreemy.ai, Undress.app and MyIMG.ai — additionally carry the note that the service is not affiliated with Clothoff, because their names are sometimes confused with ours.

Commission has no channel into the score. Weights are public and identical, tests are dated, and the order follows the number. The full text is in the Polish affiliate disclosure. If a partner changes its pricing or policy, the review is updated with a new date — not quietly rewritten.

What Clothoff AI is not: a generator, an app or a tool. The site does not accept uploads, does not process images and does not store them. Every feature described on these pages belongs to a third-party service that we tested. The Polish page czym jest platforma Clothoff AI explains the publisher role and the confusion with similarly named services.

FAQ

Questions about the method

Short answers; the Polish editorial policy is the binding document.

Who writes the reviews?

The Clothoff AI editorial team, published as an organisation. We do not use named author profiles, guest posts or user reviews; every page carries the team as author and an update date.

Do you accept payment for a higher score?

No. Six of sixteen tools carry affiliate links, marked with /go/ and a disclosure. Commission never changes weights, scores or order — tools without affiliation rank above partners where the test says so.

How often are scores updated?

Each score carries its test date (currently 12–29 August 2026). We re-test after material changes to pricing, policy or model and record the new date; old scores are not silently overwritten.

Why is privacy weighted highest?

Because every tool requires uploading a photo, and retention, deletion and payment data are where operators differ most. A privacy failure affects the user more than a quality drop.

What does “test scheduled” mean?

The tool is in the catalogue but has no completed test. We publish no estimate; the score, prices and limits appear only after a dated test — and only if legal screening passes.