How to Combine Reference Photos for a Painting: A Four-Gate Test

To combine reference photos for one painting, choose a single lead image to control viewpoint and light, then use every other photo only for named evidence such as a pose, object, marking, or texture. Match perspective and scale before correcting color, rebuild all cast shadows from one light source, and simplify the assembled scene into shared value families before adding paint marks. A seamless cutout is not yet a believable painting reference.

The common failure is democratic compositing: one photo supplies the room, another supplies a vase, and a third supplies fruit, while all three keep their own camera height, lens view, light direction, contrast, and white balance. Each piece may be accurate by itself. Together they describe incompatible realities.

What this four-pass experiment held still

We built a fictional tabletop scene from three deterministic source studies: a pale room with a left window, a blue ceramic vase photographed against gray, and three oranges photographed against a warm cloth. The final crop, object count, and intended focal point—the vase neck—stay fixed.

We changed the order and degree of reconciliation:

1. raw merge: each cutout keeps its source scale, horizon cues, light direction, and color cast;
2. geometry pass: camera height, ellipse angle, overlap, and relative scale are aligned;
3. light pass: cast shadows, value range, and color temperature follow the room;
4. painting plan: small photo artifacts disappear into larger shapes, a limited palette, and an unequal edge hierarchy.

The plate is an editorial simulation, not three photographs and not a benchmark of compositing software. Its purpose is diagnostic: each pass changes a named set of variables so you can see why clean edges alone do not create coherence.

Gate 1: appoint a lead reference

The lead reference is the scene’s physical contract. It governs:

• camera height and viewing angle;
• horizon and convergence;
• dominant light direction and softness;
• the largest light and dark masses;
• final framing and negative space.

Choose it for structure, not detail. A modest room photo with clear perspective and one readable window can be a better lead than a spectacular object close-up. Adobe’s compositing guidance groups the same compatibility checks: light and shadow, quality of light, perspective, scale, and framing.

Write one sentence before assembling anything: Lead: room photo; eye level 110 cm; soft light from upper left; vase neck is the focal point. If a later source contradicts that sentence, it is evidence to translate—not a layer to preserve literally.

Gate 2: solve viewpoint and scale before color

Geometry errors survive every palette adjustment. In the raw panel, the vase’s top ellipse is too open for the tabletop camera height, and the oranges appear enormous relative to its base. Their edges are clean, but the objects do not occupy the same space.

Start with three checks:

1. Eye level: compare which horizontal surfaces are visible. If you can see deep into the top of one vessel but barely see the tabletop, those sources imply different camera heights.
2. Scale anchor: pick one relationship you can defend, such as an orange measuring roughly one-third of the vase body’s width in this fictional design. Avoid scaling each object by feel after every move.
3. Contact and overlap: decide exactly where each object touches the table and which silhouette crosses in front. A floating base or ambiguous overlap is a spatial error, not a shadow-detail problem.

Getty’s formal-analysis material separates the picture plane from the illusion of three-dimensional space and identifies perspective plus shading as depth-making tools. For this workflow, perspective is the first gate because later shading cannot rescue a vase whose ellipse describes the wrong eye level.

Gate 3: rebuild one light, including the shadows

Do not average conflicting shadows. Delete them from the supporting references and construct new ones from the lead light.

In our raw merge, the room is lit from the upper left, the vase highlight sits on the right, and the fruit casts a short hard shadow to the left. The geometry pass makes the objects sit together, but the scene still feels assembled because the highlights and shadows disagree about where the lamp is.

Use this sequence:

• mark one light arrow outside the crop;
• place the attached shadow on the side facing away from that arrow;
• project every cast shadow in the same general direction;
• let distance and surface change softness, not direction;
• compare the darkest object shadow with the room’s existing dark family.

Adobe specifically warns that mismatched light direction, shadow direction, and hard-versus-soft light reveal a composite. For a painting plan, perfect optics are unnecessary, but contradictions must be removed. If you cannot plausibly relight a supporting object, replace its reference.

Gate 4: unify value and color, then simplify

Color matching is a finishing gate, not the foundation. Adobe’s current Match Color documentation separates luminance, color intensity, and color cast. That separation matters: pushing every source toward the same hue does not guarantee that their light-dark ranges belong together.

First reduce the assembled reference to four or five value families. In this scene they are pale wall, midtone table, dark vase, light fruit, and connected cast shadow. Then set a palette budget: muted blue, warm orange, linen gray, and one dark brown. The third panel becomes coherent because all sources now share those roles.

The final panel goes one step further. It removes tiny cutout noise, merges the fruit shadows, softens the far table edge, and keeps the crispest transition at the vase neck. That is the moment the composite becomes a painting brief rather than a polished collage.

Use masks as a rehearsal, not as the final standard

Layer masks are useful because they can hide and reveal source material without permanent erasure. Keep each donor image on a separate layer, preserve an untouched copy, and label its job: vase shape, orange arrangement, or linen fold rhythm.

But a flawless mask can make you overvalue pixel-level seams. Run three uglier checks before refining edges:

• view the composite at thumbnail size;
• convert it temporarily to grayscale;
• blur it enough that small contours disappear.

If the scene breaks under those checks, the problem is still hierarchy, geometry, or light. Feathering the mask only conceals the seam between sources; it does not reconcile their realities.

A reproducible 12-minute compatibility test

Before committing to a full painting, build a low-detail rehearsal:

1. Put the lead photo at the bottom and lock its crop.
2. Write the lead viewpoint and light direction on a note layer.
3. Place one supporting object as a non-destructive layer.
4. Hide its original background with a mask.
5. Align eye level, scale, contact point, and overlap.
6. Remove the donor shadow and sketch a new shadow from the lead light.
7. Match the object’s darkest dark and lightest light to the scene.
8. Reduce its color intensity or cast only after value fits.
9. Repeat for the next supporting object.
10. Flatten a copy into four or five large value groups.
11. Mark one hardest edge and one strongest color accent.
12. Reject any source that still needs explanation when viewed small.

The twelve minutes are a studio constraint, not a performance claim. Stop while the rehearsal is crude. Its job is to expose incompatible evidence before detail makes you reluctant to replace it.

Failure modes the final image can still hide

Matching color before perspective. A shared palette can make incompatible camera views feel temporarily related. Check ellipses, horizons, and contact points first.

Keeping every donor highlight. Highlights report light direction and surface shape. If they disagree with the lead light, repaint or remove them rather than calling them texture.

Using several “main” photos. When two references both control the pose, lighting, and framing, every correction becomes a negotiation. Name one lead and reduce the other to a specific evidence role.

Treating resolution as authority. A high-resolution donor with the wrong viewpoint is worse than a smaller image with compatible structure. Use extra references to clarify markings or texture, not to overrule the scene.

Ignoring source rights. Use your own photos, licensed sources, or clearly identified open-access images. The Smithsonian, for example, marks reusable collection assets with a CC0 designation. Availability online by itself is not permission to reuse an image.

The decision rule

Keep a supporting reference only when you can finish this sentence: I need this image for ____; it does not control ____. A useful answer is: I need this image for the vase profile; it does not control the room’s viewpoint, light, or palette.

Once the assembled reference passes the four gates, make one flattened rehearsal and use FreePainter’s photo-to-painting workflow to explore the medium. Judge whether the result preserves the single viewpoint, shared shadow family, and focal hierarchy—not whether it reproduces every donor pixel.