How to match a colour exactly instead of describing it
Give Vizard Agent the image and ask it to sample the colour rather than describing it in words. It reads the exact values out of the pixels, builds the cards and captions from those values verbatim, and then checks the colour in the finished render — because a value that was right in the design can still shift by the time it is encoded.
What is the short version?
A colour described in words comes back close. Close is fine in isolation and obvious the moment your generated card sits next to the screenshot it is supposed to blend with, because the join between two nearly identical colours is more visible than the join between two different ones.
- Go to Vizard Agent with the image.
- Ask it to sample the exact values, not match a description.
- Ask it to verify the colour in the render.
What do you need before you start?
The source image and a clear idea of what has to match what. Vizard Agent samples from pixels, so it needs the real file rather than a description or a screenshot of a screenshot, and it needs to know which element is the one everything else has to agree with.
- The source image. A screenshot, a logo, a frame.
- What must match. Background, text, an accent.
- Where the join is. That is where errors show.
- Any official values. A brand sheet beats a sample.
- What else uses the colour. Captions, cards, lower thirds.
What do you type into Vizard Agent?
Ask Vizard Agent for sampling explicitly, in those words. Left with only a description, the request quietly becomes a judgement about what "our blue" means, and every element built from that judgement ends up independently and slightly wrong.
Prompt
Variants worth knowing:
- "Sample it out of the image." The instruction that matters.
- "Those values." Used verbatim, not interpreted.
- "Check the render." Where the surprise happens.
What does Vizard Agent actually do?
Here is the pattern Vizard Agent follows across real ads that had to blend generated graphics with supplied artwork. The sampling step at the front is cheap and quick; the verification step at the end is the one that stops a subtle shift from reaching the delivered file.
- Extracts the brand's own assets where they exist.
- Analyses the supplied image for its actual values.
- Reads the exact colours out of the pixels.
- Uses those values verbatim in the graphics.
- Applies the same values to captions and accents.
- Renders the card against the real footage.
- Checks the background pixel value in the render.
- Re-renders if it shifted during encoding.
- Confirms the join is invisible on a full-size frame.
Step three is worth insisting on because a colour that looks flat rarely is. Compression, a gradient, or a subtle shadow means the value at one point is not the value at another, so Vizard Agent samples where the join will actually be rather than from the middle of the image.
Step seven is the check almost nobody runs. A value set correctly in a design can come out slightly different after encoding, and the only way to know is to look at the pixel in the finished file rather than in the source that produced it.
Step one matters when it applies. If the brand publishes its own values, those beat anything sampled from a compressed screenshot, so Vizard Agent looks for the official source before falling back to the pixels.
What does the result look like?
A card that meets the screenshot with no visible seam at all, captions in the actual brand colour rather than an approximation of it, and a confirmed pixel value from the finished file that you can point at if anybody asks whether it is right.
Vizard Agent reports the values it sampled and the values it measured back, which are the same two numbers a brand reviewer will want to compare.
When does this not work well?
Sampling gives you the colour of that image, which is not always the colour of the brand. A compressed screenshot, a photograph under coloured light, or an asset that has already been through a filter will all give you a number that is precise and wrong.
- Compressed screenshots. The value drifts from the original.
- Photographs. Lighting changes everything.
- Filtered or graded assets. Already shifted.
- Gradients. There is no single value to sample.
- Different colour spaces. The same number can look different.
How do you fix a result that came back wrong?
Say where exactly the mismatch is visible. Because Vizard Agent sampled a specific point in the image, a note like "it is wrong at the top edge" and one like "the whole card is off" point at completely different causes and lead to different fixes.
- "There is a seam at the join." Sample at the join instead.
- "The whole card is off." The source value was wrong.
- "It looks right in the file, wrong in the video." Encoding.
- "Use our official value." Given directly, no sampling.
How does Vizard Agent compare to doing it yourself?
By hand a colour gets picked once, by eye or from memory, and reused across the graphics. Each element ends up plausibly close and none of them exactly right, and the mismatch is discovered where two of them meet.
| By hand | Vizard Agent | |
|---|---|---|
| The starting point | A description or memory | Values sampled from pixels |
| Where it is sampled | The middle, if at all | Where the join will be |
| Consistency | Each element separately | One value, everywhere |
| Encoding shift | Undetected | Checked in the render |
| Evidence | "It looks right" | The measured pixel value |
Common questions
Why is describing it not enough? Because a near match is more visible than an obvious difference, so Vizard Agent samples instead.
Can it read the brand's own values? Yes. Vizard Agent looks for the official source first.
What if the image is compressed? Vizard Agent will say the sample may not be the true value.
Does it work on a video frame? Yes, though Vizard Agent will note that a frame carries the grade with it.
Can it match a gradient? Not with one value. Vizard Agent will suggest sampling a range.
Will the captions match too? Yes, if you say so. The same values are used throughout.
How does it check the render? Vizard Agent reads the pixel value straight out of the finished file.
What if it shifted anyway? Vizard Agent adjusts and re-renders until the render matches.
Can I just give it a hex? Yes, and that is the most reliable input of all.
Does colour space matter? It can. Vizard Agent will flag when the source and output differ.
What about text colour? Same approach, though Vizard Agent may argue legibility against an exact match.
Can it sample several colours? Yes. Vizard Agent can pull a whole palette out of one image.
Will it tell me the values? Yes. Vizard Agent reports them so you can reuse them next time.
Why does the seam show at all? Because the eye is far better at spotting a small difference than a large one.