Vizard Agent

How to get a generated product's proportions right from reference photos

Last updated 2026-09-10 · 8 min read

Send Vizard Agent photographs of the real product and say they are references for its proportions rather than material for the video. It compares each generated shot against them, works out what is out of scale or in the wrong place, and rebuilds those shots to the corrected proportions while the photographs themselves never appear.

What is the short version?

A generated product looks convincing and is usually the wrong size. The body is too tall, an accessory sits where nothing like it sits, a part is twice the scale it should be. Photographs of the real thing fix that, provided they are treated as measurements rather than as footage.

  1. Go to Vizard Agent with photos of the real product.
  2. Say they are for proportions only.
  3. Name the shots that look wrong.

What do you need before you start?

Photographs that show the object clearly, ideally from more than one angle, and a sentence about what they are for. That sentence matters: a supplied image is normally something to use, and these are something to measure against.

What do you type into Vizard Agent?

Say what the photos are not for. Without that line they are perfectly reasonable material for the edit, and you will find them cut into the video as product shots — which is a sensible reading of an unlabelled attachment.

Prompt

Here are photos of the real product so you can get the proportions right. Do not put these photos in the video — I am giving them to you only for the proportions and the corrections. Rework the body's proportions and where the remote sits in the shot around fifty seconds.

Variants worth knowing:

What does Vizard Agent actually do?

Here is the order Vizard Agent worked in on a real product film whose generated shots did not match the object. The comparison is done shot by shot, which is slower than one global correction and is the only thing that works.

  1. Downloads the reference photographs and tiles them together.
  2. Looks at each of the client's product images in turn.
  3. Straightens the photographs where they were shot at an angle.
  4. Opens each proportion reference individually.
  5. Compares the proportions shot by shot against the generated frames.
  6. Assesses what is wrong in each shot separately.
  7. Prepares the references in a form the regeneration can use.
  8. Checks the orientation of the new photographs before using them.
  9. Regenerates the affected shots at the corrected proportions.
  10. Extracts frames from the new shots.
  11. Checks the product's proportions closely in each of them.

Step five is the part that cannot be shortcut. Each shot has its own camera angle and its own distance, so "the product is too tall" is true differently in each one, and a single correction applied to all of them fixes some and breaks others.

Step eleven is the check that stops the correction drifting. A shot rebuilt to fix one proportion can easily arrive with a different one wrong, so the product is examined closely in each new frame rather than the shot being accepted because the original fault is gone.

Step three is a small thing that saves a wrong measurement. A photograph taken at an angle makes a straight object look tapered, so the references get straightened before anything is measured off them.

What does the result look like?

Shots where the product has the shape it actually has: the body at the right proportions, the accessory the right size and in a place it could really be, consistent from one shot to the next. Your photographs are nowhere in the video.

Somebody who owns the product recognises it, which is the only test that counts.

When does this not work well?

References fix proportions, not identity. Fine detail — a texture, a moulding line, the exact shape of a control — is not reliably reproduced from photographs, and a product whose appeal is in that detail may need real footage instead.

How do you fix a result that came back wrong?

Name the shot and name the part of the product that is wrong in it. Vizard Agent keeps your references and its own assessment of each individual shot, so a second pass corrects that one shot rather than regenerating the whole sequence at some new setting.

How does Vizard Agent compare to doing it yourself?

The usual approach is to describe the product better in words and generate again, which improves the description and not the dimensions. Supplying photographs works, but only if somebody compares them against each shot rather than glancing at the set.

By hand Vizard Agent
The fix A longer description Measured against photographs
Scope One correction for all shots Assessed shot by shot
Skewed references Used as they are Straightened first
The photos Often end up in the edit Kept out by instruction
Consistency Drifts between shots Checked against the same references

Common questions

Will my photos appear in the video? Not if you say they are references. Vizard Agent keeps them out.

How many photos does it need? More than one angle. Vizard Agent will say if depth is unclear.

Can it fix just one shot? Yes. Name it and Vizard Agent leaves the others alone.

What if the product changes between shots? Vizard Agent checks them against the same references.

Can it get fine detail right? Proportions yes; Vizard Agent is less reliable on fine texture.

What about an accessory in the wrong place? Name where it should be and Vizard Agent rebuilds the shot.

Can it use manufacturer photos? Yes, if you have the right to use them; Vizard Agent measures either way.

Does it need dimensions in numbers? No, though they help. Photographs are usually enough for Vizard Agent.

Will the style change? No. Vizard Agent regenerates at the corrected proportions only.

Can it check the whole film for me? Yes. Ask Vizard Agent for a proportions pass across every shot.

What if the photo is at an angle? Vizard Agent straightens it before measuring.

Can I supply a drawing instead? Yes, and Vizard Agent finds a dimensioned drawing better than a photo.

Will this cost a full regeneration? Only for the shots that change; Vizard Agent leaves the rest.

Why not just describe the product better? Because a description sets what the product is, and a photograph sets how big each part of it is relative to the rest.