Personas and styles

The two things that make a channel look like a channel.

The same face, and the same visual language, across everything you make. Both are built once and reused for as long as you want them.

Personas

A person your thumbnails can keep casting.

Three to ten photographs, varying angle, expression, lighting and distance. Re-Prompter writes a physical description from them, so the persona works even on an image model that takes no reference photograph at all — and rides along as reference images on models that do.

Reference-based

Works everywhere. No training, no waiting, no per-model setup.

Provider-trained

Where a provider offers identity training, Re-Prompter detects it and uses it. Where none does, it says so plainly instead of implying otherwise.

Two in one frame

Person A on the left, person B on the right, in one generation.

Consent first

A persona cannot be built without confirming you have permission to use the photographs. That check lives in the feature, not in the interface.

Styles

A look, not a subject.

Extract a style from three to ten references and Re-Prompter describes the aesthetic rather than the content: "high-contrast editorial photography, warm edge lighting, dark background, single oversized hero subject, large left-aligned condensed headline" — not "a specific person beside a specific car", because the next image will have neither.

Twelve attributes

Colour palette, contrast, lighting, composition, background, typography, visual density, saturation, texture, subject scale, framing and negative space. Each with its own switch.

A strength dial

Nought to a hundred. As it falls the weakest attributes drop out first and the palette holds longest, because the palette carries most of the look.

Measured palettes

Colours are sampled from the reference pixels and override whatever the model says. A measurement beats an opinion, and it still works when the API call fails.

See what reaches the prompt

The exact clause your style contributes is shown before you use it. No black box, no guessing why an image came out the way it did.