Reference Photos: How Someone Else's Frame Sets the Scene for Your Portrait

A reference is a photograph that sets everything except appearance: composition, light, setting, mood. Appearance comes from a separate set of your own shots. This division of roles solves a problem that neither a text description nor ordinary photo-based generation can.

Why split the roles

Describing a specific scene in words is hard. "Soft window light from the left, a side view, a blurred kitchen behind, warm palette, a three-quarter pose holding a cup" is a long phrase that still describes the scene approximately.

A photograph of that same scene contains all of it at once, without loss. A reference works as a visual brief: showing is easier than telling.

What comes from the reference

• Composition: where the figure sits in frame, the shot size, the camera angle.
• Light: direction, softness, colour temperature, presence of a rim.
• Setting: the type of room or location, the background character, how blurred it is.
• Mood and palette: the overall colour scheme.
• Pose in broad terms: body position, head tilt, the character of a gesture.

What comes from your shots

Appearance only: facial features, proportions, skin tone, hair type. Everything that makes a person recognisable, and nothing beyond that.

Clothing usually comes from the reference or the description rather than from your photographs — worth keeping in mind when choosing a reference.

Which references work best

1. One person in frame, close or medium shot. Group scenes complicate the task and often produce blending.
2. A clear composition without complex occlusion: the figure reads as a whole.
3. Natural light with a clear direction: models reproduce it more accurately than elaborate studio setups.
4. A simple pose without extreme angles or gestures near the face.
5. Good quality: a soft reference gives a soft result.

Which references give poor results

• Frames with strong low or high angles: transferring them to another face distorts proportions.
• Scenes with complex interaction between hands and objects.
• Frames where the face is partly obscured — there is nothing for the model to align to.
• Images with large text or logos.
• Frames whose point is a specific recognisable person: transferring a scene is not transferring an identity.

Aspect ratio

The frame format is usually inherited from the reference — logically, since composition and format are inseparable. If you need a different format, it is better to find a reference in it than to crop the result.

Boundaries: whose frame is this

A reference is someone else's photograph, and that deserves saying. Using another's shot as a guide to light and composition is ordinary practice, familiar to any photographer: compositional decisions are not anybody's property.

But reproducing a recognisable authored frame in full, especially commercially, is a different matter. A practical boundary: take principles from the reference (light, shot size, mood) rather than the specific frame with all its particulars.

And separately: if the reference contains a recognisable person, the result must not create the impression that it is them.

Additional parameters

Some scenarios let you specify build — height and weight. This helps when the reference shows a full figure: without those numbers the model picks default proportions, and the result may not match reality.

If the parameters are omitted nothing breaks — the model simply uses the most probable option. They are worth supplying when the discrepancy would be noticeable.