How we test and what we publish

Every image model on this site runs the same six prompts, split across three tiers: SFW, erotic and explicit. All of them run on the same rig, with no cherry-picking and no retouching. The rig and the run date are printed on each model page, so you can see what produced the result and how old it is.

The prompts are chosen to break things. They stack hands, readable text on fabric and on skin, mirror geometry, genuine expressions and several distinct characters in one frame. A prompt that every model passes would tell you nothing about which one to reach for.

The three tiers exist so that a model which refuses adult prompts outright still gets a fair, comparable test on the SFW tier instead of a blank page.

Each prompt has a fixed checklist, and every check is graded pass, partial or fail. Partial exists on purpose. It is the right grade far more often than a clean pass, and a rubric that cannot say partial produces the table where one model wins every row, which reads as marketing rather than testing.

A model page only shows images that model produced. We never borrow another model's render to fill a gap. Each tier is marked complete, not supported or pending. Not supported means the model's own filters or terms block that tier, which is a fact about the model and does not change until the model does.

Explicit output never appears on this site. The explicit tier is still run as a capability test, and the model page records whether the model could do it, but the page shows the same locked placeholder for every model. The renders themselves are never shown here, blurred or otherwise.

Prices and plan limits are copied by hand from the vendor's own published page, and each figure carries the date it was last verified. Directory entries show the date they were checked and link to the tool's own site, so you can confirm a claim at the source.

When a fact comes from a vendor's page and we have not tested it ourselves, the page says so. A vendor's claim is labelled as the vendor's claim.

Some links on this site are affiliate links. The disclosure page explains which ones and how they are handled. GenLovers reviews and ranks tools on what they do in testing, including tools that competed with Honeybot, the AI companion platform its author co-founded and closed in 2026.

When a roundup includes a product we build, the roundup says so and does not rank it above the others on that basis.

Every guide, model page and review shows when it was last updated, and the sitemap carries the same date. When a page changes materially, that date changes with it.

If you find a wrong price, a stale claim or a result you can reproduce differently, write to us through the contact page. We fix the page and update its date.