Our test character

Our test character is an AI-generated adult who does not exist. We made it on 2026-10-21 with Soul 2.0 on Higgsfield, for 0.48 credits (our charge). It is a test subject for the method on the How we test page, nothing more.

Who the test character is, and who it is not

As of 2026-10-21, the test character is an original, fictional adult woman, about 29 years old according to the prompt. We built it from a text description only; no photo of a real person went into it. This is the full prompt, verbatim from our log:

Front-facing waist-up portrait of an original fictional adult woman, about 29 years old, shoulder-length curly copper-auburn hair, light freckles, round tortoiseshell glasses, warm curious smile, wearing an olive-green bomber jacket over a plain white t-shirt, small silver hoop earrings. Clean light-grey seamless studio background, soft even daylight, natural skin texture, candid creator vibe. Ordinary person, not a celebrity.

We did not run a face-matching check against public photos: not tested. It has no name yet, so the site calls it “our test character”.

Why our examples use fictional characters

Effect and trend pages need a face to demonstrate on, and since 2026-10-21 every example on the site with one person uses this one. Couple photos and hug videos add the second character below. We do not use other people’s faces, and we do not show our staff. In each comparison every model gets the same input and prompt, so differences show up in the output.

Because the face belongs to no one, we can publish a bad result as it came out. In the Christmas polaroid run (Nano Banana Pro, 2026-10-21, 2 credits, 1 try), the hair came out darker and straighter than the reference.

How it was made: models, date, credits

Both steps ran on 2026-10-21 on a Higgsfield Plus plan account, and together they charged 2.48 credits.

1. Soul 2.0: the prompt above, 4 variants, 0.48 credits, 1 try, aspect ratio 3:4. The job record names the model text2image_soul_v2. We chose variant 3 as the canonical image.
2. Nano Banana Pro: one run with the canonical image as reference, set on a city street at golden hour, 2 credits, 1 try. Our log marks identity as held. The job record labeled the model nano_banana_2; we draw no conclusion from it.

In our six test sessions (2026-10-21 to 2026-10-27), 43 image runs took the canonical image directly as their reference. Four of them also took the second character. Another, H9-dmg, turned it into a synthetic, scratched “old photo” for our old photo restoration test. Video runs of the character started from it or from stills made with it. These were G1 and G3 from the first session, H0 from 2026-10-26, and a restored portrait and a couple photo from 2026-10-27. H0 went to Seedance 2.0, Hailuo 2.3 and Veo 3.1 with one prompt. The exception is the eight Soul 2.0 preset runs of 2026-10-27. A Soul style preset cannot be combined with a reference image, so our Soul 2.0 preset test describes the character in the prompt.

Three subjects show no person, and we generated each from text. One is the fictional dog in our pet dance test, another the mannequin clips behind our motion control test. The third is H6a, a rainy street with no person (2026-10-27).

Our second test character, for scenes with two people

On 2026-10-27 we added a fictional adult man who reads 33 to 38. He is the partner in our couple photo and hug video tests, so no real person is needed. Soul 2.0 made four variants of him from one text prompt, with no reference image, for 0.12 credits each. We approved variant d the same day: it reads clearly adult, has no render artifacts and is framed like our first character. We rejected a (frame-corner marks, an unrequested neck tattoo), b (reads younger, retouched) and c (a mark at the right eyebrow).

AI-generated test image of our second fictional test character. Soul 2.0 via Higgsfield, 2026-10-27, 0.12 credits (variant d of 4), 1 try.

He has close-cropped black hair with a light-gray patch on top and receding temples. He also has a full black beard with gray flecks and wears a mustard ribbed crew-neck sweater. The mole the prompt asked for did not come out clearly, so it is not a marker. Face-matching check: not tested.

What we never use them for

Under our policy as of 2026-10-21, we never present either character as a real customer, tester or expert. No quotes, ratings or reviews are attached to them. Every image of them carries a caption that labels it AI-generated and gives the model, date, credits and tries.

Rules for real photos live on the content policy page. In short: use your own photo or a photo of an adult who agreed, and expect no how-to for other people’s faces here.