UncensoredMethod
Each tool receives seven readings out of ten. The lab score is a weighted average scaled to 100. Nothing is curved. No score is massaged after the fact to clean up a ranking.
The seven readings
| Realism | 22% | Skin, hands, lighting and anatomy on the same prompt set. Photoreal when asked, stylised when asked. |
| Prompt control | 17% | How much of the result you steer: model choice, reference images, LoRAs, negative prompts, seeds, inpainting. |
| Consistency | 15% | Same subject, four scenes, then video. Does the face, body and styling survive, or does each render produce a stranger? |
| Freedom | 18% | What the tool lets you make. Explicit content on by default, opt-in, or blocked, and how often moderation false-positives. |
| Video | 10% | Whether it does video at all, clip length, motion quality, and whether video is explicit-capable. |
| Usability | 6% | Time from sign-up to first good output, queue times, and whether the interface gets in the way. |
| Value | 12% | What a month of real use costs against what you get: credits, unlimited tiers, watermarks and free allowance. |
Score bands
83 83 and up: best in class72 72 to 82: good58 58 to 71: usable with caveats40 below 58: not recommended
The bench protocol
The same prompt set hits every tool, run on a paid account: twelve still-image prompts covering photoreal, stylised and anime, at three explicitness levels; four consistency prompts that request the same subject across different scenes; and, where available, four short video prompts. Outputs are kept, timed, and scored blind by two people. Refusals and silent edits hurt freedom; hands, skin and lighting boost realism; following the prompt exactly boosts control.
Scores tagged provisional have not yet hit the bench. They are assembled from documented features, current pricing and published third-party reviews, and the page states this. Once the bench run completes, the score updates and the changelog notes what shifted.
What we will not score
Any tool allowing content with minors, or non-consensual imagery of real people, is excluded, not linked, and reported. Tools marketing face-swapping of real people are excluded.