Choose neutral gray when value is uncertain, warm earth when you need a fast drawing and shared midtone, a restrained complement only when some of it will remain active, and local color when the major color relationships are the risk. There is no universally correct hue. First decide what the layer must solve; then decide whether later paint will cover it, break over it, or transmit it.
That distinction matters because an underpainting is not merely a colored background. The National Gallery of Art defines it as an initial paint layer that begins to establish shapes and values. Getty’s Art & Architecture Thesaurus similarly describes a preliminary layer where drawing, composition, and tonal values may be worked out, or a single-color area prepared for later glazes.
First ask how much of the start will survive
Color matters differently in three finishes. If later paint will be opaque and complete, choose the underpainting mainly for drawing, value, and speed; the hue may guide the process without remaining visible. If you plan broken coverage or scumbling, select a restrained color that can appear between later marks without taking over. If you will glaze, the lower layer participates optically, so a digital preview is only a plan: test the actual paint, binder, surface, and drying state on a small card.
This is a decision diagram, not a pigment measurement. The National Gallery distinguishes glazing—transparent paint that influences the layer below—from scumbling, where opaque or semi-opaque dry paint only partly covers what is underneath. Those different layer behaviors are why “warm or cool?” is the second question, not the first.
What this comparison can—and cannot—test
Our cover is an editorial simulation, not four photographs of physical paintings. We held the fictional sitter, three-quarter pose, expression, crop, drawing, blouse, background, and upper-left side light stable. We changed only the instruction for the starting layer:
- neutral gray tonal block-in;
- transparent warm earth wash with lifted lights;
- muted green complementary start beneath the flesh zones;
- simplified local-color masses.
We stopped every panel before a finished paint layer. That lets us compare what each start makes easy to judge. It does not prove how a particular pigment will dry, cover, stain, or show through. Opacity, binder, surface, layer thickness, and later handling can change the physical result. Use the plate as a planning test, then make a small material study with your actual paint.
Neutral gray: when value is the main uncertainty
The gray panel makes the light pattern easiest to audit. Hair, face shadow, neck, blouse, and background can be compared without hue competing for attention. Use this start when likeness depends on the turn of the head, when the reference has confusing color casts, or when you keep mixing attractive colors that do not share a believable light structure.
Reduce the source to three or five value families first. Place the darkest connected mass, the lightest face plane, and the background value before describing eyes, lips, or hair strands. If the portrait reads at thumbnail size, the start has done its job.
Failure mode: a fully modeled grayscale portrait can become a separate finished picture that you are afraid to disturb. Stop when the large planes are reliable. Do not polish every eyelid before color begins.
Warm earth: when drawing speed and a shared midtone matter
The burnt-sienna panel carries the drawing with one family of marks. Thin paint can establish placement; a rag or brush can lift the light; denser passages can hold the shadow. In our comparison, that economy makes the head construction readable without asking for a complete color plan.
The choice also has historical precedent without being a universal rule. The National Gallery of Art’s technical account of Hobbema’s A Farm in the Sunlight records a dark reddish-brown ground and a light-brown foreground underpainting that served as a midtone. That is evidence of one working method, not proof that every landscape or portrait wants brown.
Choose a warm earth start when the subject already contains wood, warm skin, autumn color, or ochre fabric—and when small uncovered gaps would belong in the final harmony.
Failure mode: a dark or overly opaque earth layer can make every later light an expensive correction. Keep the wash lighter than the deepest final dark and reserve enough surface for clean lights.
Complementary start: when controlled contrast should stay active
The muted green panel is the most assertive option. Against warmer flesh and ochre clothing, the start makes missed gaps and thin passages visibly cool. Adobe’s current complement tool documentation is a useful reminder that “complement” is a defined color relationship, not a synonym for any dramatic opposite-looking hue.
This approach also has a specific historical example: the National Gallery records green underpainting beneath flesh tones in Saint Matthew. The lesson is not “paint every face green.” It is that a contrasting layer can participate in later color when the layering method is designed for it.
Use a complement only if you can name where it may remain visible: between warm strokes, along a turning cheek, inside a shadow transition, or around a warm focal object. If the finish will be uniformly opaque, the contrast may affect the process more than the visible result.
Failure mode: a saturated green start can contaminate thin flesh mixtures and force the whole portrait into a color-correction exercise. Mute it, test coverage, and keep the value close to the intended middle family.
Local color: when the big color relationships are the risk
The local-color panel answers a different question: does the warm face separate from the dark hair, ochre blouse, and neutral background before detail? It is the strongest start when the reference contains several large color masses whose relative temperature and saturation matter more than a monochrome drawing.
Block each major region with one average mixture. Do not sample every pixel. The point is to compare families: is the background quieter than the face, is the blouse warmer than the light, and is the hair dark enough to frame the profile without becoming a black cutout?
Failure mode: local-color block-in can turn into premature rendering. If you are already mixing five cheek colors, return to one light family and one shadow family.
A reproducible four-card test
Before committing a large surface, make four small studies from the same reference. Keep the crop, drawing, palette, brush size, and time limit fixed. Change only the start. Photograph all four under the same light, then review them in this order:
- Which card preserves the likeness or composition with the fewest corrections?
- Which one makes the main light direction obvious?
- Which one gives later colors a useful neighbor rather than a color cast to fight?
- Where would you intentionally leave the first layer visible?
- Which start tempts you into detail before the large masses are settled?
Pick the card that removes the highest-cost uncertainty. For a portrait, that may be value and geometry. For a warm interior, it may be a connected earth midtone. For a color-driven still life, it may be local relationships. For a loose painting built from broken coverage, a restrained complement may earn its place.
Before scaling up, add one coverage strip across every card: one opaque stroke, one dry broken stroke, and one thin transparent passage made with the same later color. Compare the strips only after they reach the working state your real process requires. This separates two decisions that are often confused: whether the start helps you organize the picture, and whether its hue helps the final layer stack.
Then carry the chosen study into a single full-size attempt. If you want to test the warm-earth or neutral route as a painterly derivative before using physical materials, start with the FreePainter oil-painting workspace and keep the same source image across versions. Judge structure first; surface effects come later.
References
The source notes below establish what an underpainting does and document several real techniques. Our four-panel comparison and decision matrix are original editorial simulations; they are not reconstructions of the cited artworks or measurements of pigment performance.