The best reference photo for a portrait painting preserves four kinds of evidence: believable facial geometry, one readable light direction, room around the head and shoulders, and sharp identity anchors—especially the eyes. If one of those is missing, a painting tool must invent information before it can interpret the subject.
Resolution is not the first test. A large close-up can record every pore while distorting the nose-to-ear relationship. A beautifully exposed headshot can flatten the face until the cheek, jaw, and eye socket no longer separate. Choose the source that explains the head, not the source that merely contains the most pixels.
What we held still
Our cover is an editorial simulation, not four photographs of a real sitter. We held the fictional sitter’s identity, age, expression, copper bob, teal blouse, cream background, and three-quarter pose stable. We changed only three source variables:
- camera distance and the resulting perspective;
- how much room the crop leaves around the head and shoulders;
- the number, color, and direction of visible light sources.
The four panels are not a camera or model leaderboard. They are a source-selection test. The question is not “Which photo is prettiest?” It is “Which photo gives a painting enough stable evidence to preserve likeness while simplifying detail?”
Test A: reject the too-close phone photo
The upper-left panel looks detailed because the face fills the frame. It is also the weakest primary reference. The nearby camera enlarges the closest feature, compresses the apparent distance to the ears, and clips the hair and shoulder before a painter has chosen a crop.
Adobe Research describes the familiar selfie problem in the same practical terms: a lens held extremely close can make the nose appear larger and the ears smaller. That is a perspective problem, not something that more brush texture can solve.
Check geometry before style. Compare:
- visible cheek width on the near and far sides;
- nose length relative to mouth width;
- ear size and placement;
- jaw angle from ear to chin;
- the amount of forehead and hair mass left in frame.
If the photo came from an arm’s-length front camera, take another from farther away. A longer camera setting can help frame the head while you step back, but distance is the useful action. Do not ask a painting pass to “make the face more natural” after accepting warped evidence.
Test B: keep flat light as an identity backup
The upper-right panel fixes distance and crop. The eyes, jaw, hair silhouette, and shoulder line are available. Its problem is subtler: broad frontal light compresses the light-to-shadow separation that explains form.
This photo can still be useful. Treat it as an identity backup for eye color, hairline, clothing, and proportions. Do not make it the only source if the final portrait needs convincing volume. When the forehead, near cheek, far cheek, and jaw all sit in nearly the same value family, the painting has to invent where the planes turn.
The Getty’s formal-analysis guide defines value as how light or dark a color appears and explains that shading helps imply three-dimensional space. For a portrait workflow, that means the source should reveal a connected light shape and a connected shadow shape before it reveals freckles.
Try the thumbnail test: reduce the photo until eyelashes disappear. If the head still reads as a solid form, the value structure is doing useful work. If only the outline remains, keep the photo for identity facts and find a second lighting reference.
Test C: do not paint a color argument
The lower-left panel has workable geometry and crop, but warm overhead light and cool screen light divide the face into conflicting color systems. The effect can be expressive when it is intentional. It is unreliable when you want a neutral skin-color reference or a reproducible photo-to-painting comparison.
Mixed light creates two separate questions: is a blue cheek actually cooler local color, or merely screen light? Is the orange temple skin color, reflected wall color, or a warm bulb? A painting model may answer differently in different regions, producing a patchwork face.
Use this panel only as a color or mood backup. For the primary source, turn off competing lamps, move away from colored screens, and look for one dominant light. Adobe’s portrait-lighting guidance starts with the same observation: identify where the main light sources come from before making the portrait.
Test D: choose one readable side light
The lower-right panel passes all four checks. The camera is far enough away for believable proportions. Hair, chin, neck, and shoulders remain available for a deliberate crop. The eyes are sharp. One side-window direction creates a light family and a shadow family without erasing the far eye.
This does not make side light universally “best.” It makes the source legible. The Getty’s Edgar Allan Poe teaching sheet shows how light and shadow can establish both emphasis and character in a portrait. Adobe likewise notes that changing a light’s angle changes the mood and facial description. The practical requirement is not a named lighting setup; it is a direction you can point to.
Before accepting the photo, ask:
- Can I trace one main path from light source to forehead, nose, cheek, and neck?
- Does the shadow eye still contain enough information to place it?
- Are the brightest skin areas below clipping and the darkest areas above featureless black?
- Does the crop leave enough room to move the portrait slightly up, down, left, or right?
If all four answers are yes, the source is ready for medium decisions.
Use a primary source and backups, not a collage of compromises
One photo should control geometry, expression, and the main light. Additional photos may clarify hair color, clothing, or a hidden ear. They should not vote on the face.
When two sources disagree, keep the primary source’s head turn and light map. Importing the left eye from one photo, the mouth from another, and the jaw from a third creates a person who never occupied one camera position or one light field. That composite may be imaginative, but it is no longer a controlled likeness workflow.
This test also has limits. It does not judge whether the expression feels characteristic of the sitter, whether you have permission to use the image, or whether the chosen crop suits the final canvas. It does not prove that one-light portraits are artistically superior. It only identifies which source makes geometry, form, and later simplification easier to defend.
The useful next step
Retake before you prompt. Stand farther back, keep the full hair and shoulders, place the sitter near one window, turn off competing colored lights, and focus on the nearer eye. Save one neutral identity backup, but mark a single photo as the geometry-and-light authority.
Then use the deepest matching workflow: turn the portrait reference into a painting while keeping the head turn, expression, eye spacing, and connected shadow shape fixed. Change the medium only after those anchors survive the first result.