How to design one location and put your character in it

Last updated 2026-08-30 · 7 min read

What is the short version?

A character filmed or drawn against a blank background needs somewhere to be. Designing one location properly and using it throughout is what turns a series of clips into a place, and it is cheaper than generating a new background every time.

1. Go to Vizard Agent and describe the location in detail.
2. Upload the character clip, shot against a plain background.
3. Say it is a single set that will be used throughout.

What do you need before you start?

A written description and a clean plate. Vizard Agent generates the set from your words and matches its scale and palette to a frame of the character, so the more concretely you describe the space, the less it has to invent.

• The description. Objects, light, mood, time of day.
• The character clip. On white, green or any flat colour.
• The frame shape. Vertical, square or wide.
• The palette. Warm, cold, or matched to the character.
• The reuse. Say it is one set for many clips.

What do you type into Vizard Agent?

Describe the room the way a set dresser would. Vizard Agent generates from specifics — the shelves, the light source, what is on the table — and a list of concrete objects produces a far more usable set than an adjective like "cosy" on its own.

Prompt

Variants worth knowing:

• Matched to the character. Scale and palette taken from a reference frame.
• Several camera angles. The same room, seen differently.
• Empty first. The set delivered on its own for reuse.

What does Vizard Agent actually do?

Here is the order Vizard Agent worked in on a real single-location build. The stretch from the twelfth step is the honest part: the background removal did not preserve transparency, so it read the operation log to find out why rather than guessing.

1. Checks the clip's duration and extracts frames to see the character and the composition.
2. Reviews the generation options and the available output formats.
3. Prepares a character frame as a reference so the set matches its scale and palette.
4. Generates the finished cinematic background to the written description.
5. Reviews the generated set for composition, light and unwanted extra characters.
6. Removes the flat background from the whole source clip and builds a check composite.
7. Inspects the character's edges and how well it separated from its original background.
8. Diagnoses the output format, extracts the alpha channel separately, and reads the operation log to find why the transparency was not preserved.
9. Compares the original and separated frames, tests a chroma key that keeps the character's white detail, and tunes the key to the footage's actual white value.

Step 9 is the specific problem this kind of job always hits. A white character on a white background gives the key almost nothing to work with, so the threshold has to be set from the measured colour rather than from an assumed pure white.

What does the result look like?

From the run this page is written from, probed on the delivered file: 1080x1920, H.264, 25fps, 126.28 seconds, AAC audio. Two minutes of the character in a single generated room — warm evening light, the described objects on the shelves behind — with the character's own light details intact rather than keyed away.

The set exists as its own asset afterwards. That is the reason to do it this way: the next clip goes into the same room without generating anything, which is what makes a series look like it happens somewhere.

When does this not work well?

Separating a character from a plain background and dropping it into a generated set has two separate failure points, and they compound each other. These are the ones worth understanding before you commit a whole series to a single generated location.

• Matching colours defeat the key. White fur on white loses detail first.
• Soft edges go crunchy. Hair and fur are the hardest thing to separate.
• Generated sets vary. Regenerating gives a slightly different room.
• Lighting has to agree. A character lit flat sits oddly in a moody set.
• Descriptions get interpreted. Vague words produce a room you did not picture.

How do you fix a result that came back wrong?

Say whether the set or the separation is at fault, because the fixes are different. Vizard Agent keeps the generated background, the reference frame, all the alpha tests and the measured key values, so one side can be redone without touching the other.

• "His edges are chewed up." The key is re-tuned to the measured background colour.
• "The room is too busy." Regenerated from the same description, simplified.
• "He looks pasted on." Relit and graded toward the set's own light.

How does Vizard Agent compare to doing it yourself?

By hand this means commissioning or generating a background, keying the character, and then discovering that the key ate the white parts of them. The diagnosis — why the alpha did not survive — is the step that stops most people.

Common questions

How detailed should my description be? Very. Vizard Agent generates from concrete objects and light, not from adjectives.

Does my character need a green screen? No. Vizard Agent keys a flat background of any colour, though white on white is the hardest case.

Will the set match my character? Yes. Vizard Agent uses a character frame as the reference for scale and palette.

Can I reuse the location? Yes, and you should. Vizard Agent delivers the generated set as an asset you keep.

What if the character has white or pale details? Vizard Agent measures the actual background colour so the key does not eat them.

Why read the operation log instead of just trying again? Because a transparency that silently did not save looks identical to a key that failed, and retrying blind fixes neither. Finding the actual cause is faster than three more attempts.

Can I get several angles of the same room? Yes. Vizard Agent generates further angles from the same description and reference frame.

Does this work for a real person on camera? Yes, though Vizard Agent treats a person against a plain wall as a slightly different keying problem.

Can I have the empty set on its own? Yes. Ask and Vizard Agent delivers the background as a separate asset.

How many clips can share one set? As many as you like, and that is the economic argument for the approach. Vizard Agent generates the room once and composites into it thereafter, so the tenth clip costs no more than the second.

Does Vizard Agent check the composite? Yes. It reviews the character's edges against the set before delivery.