Quick answer

If the same object looks like a different color in shot 1, 2 and 3, the problem is usually not the object description — it's that you changed too many variables at once. Fix it by treating the clip as one fixed object, one fixed camera, one fixed background, and three lighting states. Write the object sentence once, paste it word-for-word into all three shots, and let only the light description change. Then review each shot paused, and re-generate only the shot that drifted.

That is the whole method. The rest of this article is the detail that makes it hold up.

Why the mug changes color between shots

Picture a plain unbranded ceramic mug on a wooden table:

Shot 1: warm desk lamp from the left
Shot 2: cool window light from the right
Shot 3: backlit, mug silhouetted

The mug's shape and framing stay fixed. Only the light changes. That is a deliberate, readable sequence — and it is also exactly the setup where color drift shows up, because a warm lamp genuinely does make a neutral ceramic look amber, and cool window light genuinely makes it look blue-grey.

So the first thing to separate is intended color shift from unintended color shift.

What you see	Likely cause	What to do
Mug reads amber in shot 1, blue-grey in shot 2, dark in shot 3 — but the shape and position match	Lighting is doing its job	Keep it. This is the sequence working.
Mug reads amber in shot 1, then a different amber in shot 2, and a different object shape in shot 3	Description, camera or background changed between shots	Re-shoot the drifting shot only
Mug changes hue mid-shot, or the table color changes too	Background or framing moved	Re-shoot that shot with the background locked

This distinction matters because "consistent color" in a three-shot lighting sequence does not mean the pixel values are identical. It means a viewer still reads the object as the same object across all three shots. That is a subjective judgment you make by eye, not a number anyone can hand you.

Step-by-step workflow

1. Pick one simple object with a single dominant color

Skip anything with fine print, gradients, patterns, or reflective chrome. A plain ceramic mug, a matte water bottle, a single piece of fruit, a solid-color notebook — all good. Text, logos and busy patterns give the tool more to reinterpret, and they also make it harder for you to judge whether the color actually drifted.

One dominant color also gives you a clean reference. If the object is "a matte terracotta mug," you have one hue to check. If it's a striped mug, you have five.

2. Lock camera distance, angle and object position before generating anything

Decide these once and write them down as fixed text:

Camera height and distance (e.g. "eye level, about 60 cm from the mug")
Angle (e.g. "slightly above, three-quarter view")
Where the object sits in frame (e.g. "mug centered, table edge visible at the bottom")
Background (e.g. "plain wooden table, unbroken, no other objects")

Then reuse that exact wording in every shot. If you re-describe the camera in different words each time, you have introduced a variable you didn't mean to introduce.

3. Write each shot as a light-only change

This is the core move. Keep the object and camera sentences word-for-word identical across all three shots, and change only the lighting sentence.

4. Keep the background plain and unchanged

A plain background is your control. If the table color shifts between shots, your eye will read the mug as having shifted too, even when it hasn't. Avoid backgrounds with other objects, strong texture, or a color close to your subject's color.

5. Review by pausing each shot

Play the clip, then pause on a representative frame in each shot. Ask:

Does the hue still read as the same object?
Does the silhouette still read as the same object?
Did the background stay put?

If one shot fails, re-generate that shot. Do not re-generate the whole clip — you'll lose the two shots that were working.

Example input

Here is a copyable starting point. Treat it as a proposal you adapt, not a formula that guarantees a result.

text

15-second vertical video, 3 shots, one continuous subject.

FIXED SUBJECT (use this sentence unchanged in all three shots):
A plain unbranded matte terracotta ceramic mug, no text, no logo,
no pattern, sitting centered on a plain wooden table.

FIXED CAMERA (use this sentence unchanged in all three shots):
Eye level, slightly above, three-quarter view, mug centered in frame,
table edge visible at the bottom of the frame, plain unbroken wooden
table background, no other objects.

SHOT 1: Warm desk lamp from the left, soft shadows falling right.
SHOT 2: Cool daylight from a window on the right, soft shadows falling left.
SHOT 3: Strong backlight behind the mug, mug mostly silhouetted, table
still visible.

No text, no logos, no on-screen graphics. Vertical framing.

Two things to notice. First, the fixed blocks are literally repeated — that repetition is the technique. Second, the lighting lines describe direction and quality of light, not a color grade. "Warm desk lamp from the left" gives the tool a physical situation to render; "make it orange" gives it a filter instruction that may or may not respect your object.

If you want to build the clip inside a workspace that handles video generation alongside the rest of your assets, the AI product video generator page covers the broader capability. Keep in mind that a page describing a capability doesn't tell you which specific models are wired into it — check the current product itself for that.

Output review checklist

Run this before you export. It takes about a minute.

Three shots, 15 seconds total, vertical. Check the actual export, not the preview.
No text, no logos, no screenshots, no invented product UI. AI video tools will sometimes hallucinate a label onto a "plain" object. Look closely at the mug's surface.
Only lighting, background or framing changed. If the object's shape, material or position moved, that's drift.
Object description identical across shots. Compare your own prompt text side by side.
Background unchanged. Same table, same color, same visible area.
Silhouette reads as the same object in shot 3. Backlighting is where shape drift hides.
At most 30 spoken words. If you're adding voiceover, count it. Short clips punish long scripts.
Paused frames look like one object, three moods. This is the subjective call — make it yourself.

Common mistakes

Rewriting the object description "for variety." Paraphrasing feels harmless. It isn't. "A terracotta mug" and "a clay-colored coffee cup" may render as different objects with different hues. Copy and paste.

Changing the camera to make a shot more interesting. A new angle means a new generation with new variables. If you want variety, get it from the light.

Using a background with other objects. A second object gives the tool a second thing to reinterpret, and it gives your eye a second color reference that competes with the mug.

Judging color from the moving preview. Motion hides drift. Pause.

Re-generating the entire clip when one shot fails. Expensive and unnecessary. Re-shoot the one shot.

Expecting the mug to be pixel-identical across all three shots. It shouldn't be. Warm lamp light and cool window light produce different colors on the same object. That's physics, and it's the point of the sequence.

Adding a color grade on top. If you're grading, grade all three shots with the same settings, and check the result paused. A grade applied per-shot will undo the consistency you just built.

Troubleshooting specific failures

The mug changes hue mid-shot. Usually a framing or background change. Lock the background sentence and re-shoot.

Shot 3 loses the mug entirely. Backlighting plus a dark object can collapse into a silhouette with no readable color. That may be fine — a silhouette still reads as the same object if the outline matches. If it doesn't read, soften the backlight description so some surface is still lit.

Two shots match, one doesn't. Re-shoot the odd one. Compare its prompt text against the other two before you do; the difference is usually visible in the wording.

The object looks right but the table changed. Your background description wasn't specific enough. Name the surface and say "plain, unbroken."

A label appeared on the "plain" mug. Re-shoot and re-state "no text, no logo, no pattern." This is a common failure with plain objects.

FAQ

Does the object description have to be identical word-for-word, or just similar? Word-for-word is the safer default. You can't know how a given tool weighs each word, so removing the ambiguity is the cheapest insurance available.

Should I use the same seed across shots? Seed controls and parameter behavior vary by tool, and this article can't tell you what any specific model does with them. Check your tool's own documentation. The method above works without touching those settings.

How do I know if the color drift is a problem or just the lighting? Pause each shot. If the shape and position match and the hue difference is explainable by the light direction, it's the lighting. If the shape moved or the hue changed in a way the light doesn't explain, it's drift.

Can I use this for a product with a logo? The brief here is deliberately logo-free. Logos and fine print are the hardest things to keep stable, and they also raise questions about what you're allowed to depict. Start with a plain object.

What if I need more than three shots? The method scales, but each added shot adds a chance to drift. If you're going longer than 15 seconds, consider whether the clip still needs to read as one continuous subject.

Do I need to pay for anything to try this? That depends entirely on the tool you use. Check pricing for the current details rather than assuming a free tier exists.

The short version

Fix the object, fix the camera, fix the background, change only the light. Copy the fixed sentences instead of rewriting them. Pause to review, and re-shoot one shot instead of three. The mug should look like the same mug in three different rooms — not like three different mugs.

How to Keep a Subject's Color Consistent Across Three Shots in a Short AI Video