Choose a campaign palette as one dominant colour, one accent, and two neutrals, then test it in the actual light the work will be shot or generated in before committing. Palettes fail in production for a boring reason: they are chosen as swatches on a screen and then applied to a world with real light, where a colour that read as warm sand becomes a muddy beige under overcast daylight.
Setup
You need the brand board constraints, the placements the campaign runs in, and one sample of the physical product or environment if either exists. Palettes for physical products are constrained by the product; deciding otherwise means the product will look wrong in its own campaign.
The structure that works
One dominant. Roughly 60 to 70% of the frame across the campaign. This is the colour people will remember.
One accent. Under 10%, used where attention should go. Accents used generously stop being accents.
Two neutrals. One light, one dark, doing the structural work. Most of the sophistication in a palette lives here, not in the accent.
Four colours total, in stated proportion. Palettes presented as five equal swatches get used as five equal colours, and the result reads as decoration rather than direction.
Step by step
1. Start from the fixed points. Product colour, packaging, brand colours, and anything you cannot change. Palettes built without these produce work the product cannot appear in.
2. Choose the dominant against the medium. A palette for feed video behaves differently from one for print. Saturated colours survive compression better; subtle mid-tone distinctions do not.
3. Pick the accent last and use it least. The accent is the campaign's punctuation, and the temptation is always to use more of it than the design needs.
4. Test in light, not in a swatch grid. This is the step that separates palettes that survive from palettes that get quietly abandoned in week two. Generate the same scene in the palette under three light conditions before signing off: morning window light, overcast, and artificial interior. On 8frame that is three reference-conditioned images at 6 credits each on Seedream 5, roughly $0.18 total at pack rate.
5. Check it at thumbnail size. Shrink the test frames to 200 pixels wide. If the dominant and accent stop being distinguishable, the palette will not work in a feed.
6. Check accessibility on anything with text. Contrast ratios for overlaid copy, and a check that the palette does not rely on a red and green distinction to carry meaning.
Why palettes fail
Chosen on a calibrated monitor, viewed on a phone at half brightness. Test on a phone, outdoors, at least once.
Ignored platform compression. Meta and TikTok recompress aggressively, and the first casualty is subtle gradation in mid-tones. Palettes built on close mid-tone relationships come back as mush.
Fought the product. A campaign palette that clashes with the product's own colour means every frame containing the product is a compromise.
Too many colours. Six-colour palettes are two palettes, and the work will split into two looks.
Where generation helps
The useful application is testing, not choosing. Generating the same composition in three palettes under the same light shows you a comparison a swatch grid cannot: how the colours behave against skin, against the product, in shadow.
Practical cost: three palettes at three light conditions is nine images, 54 credits on Seedream 5 or 189 on Nano Banana Pro, roughly $0.54 to $1.89 at pack rate, well inside a $19 Starter month. Doing this before a shoot is the cheapest insurance in the process, because a palette decision reversed after production is a reshoot.
FAQ
How many colours should a campaign palette have? Four: dominant, accent, and two neutrals, with proportions stated.
Should the campaign palette match the brand palette exactly? It should live inside the brand's rules while being narrower. A campaign that uses the full brand palette has no palette of its own.
How do I test a palette before shooting? Generate the same scene in each candidate under the light conditions you will actually shoot in, then compare at both full and thumbnail size.
Palette tests are most useful side by side, at real size and at thumbnail, on the same surface. The 8frame canvas is free and unlimited, and generation is paid from $19/month. See also what is color grading.