How to turn a talking-head clip into a paper collage
Upload a short clip and give Vizard Agent a four-colour palette. It removes the background, turns you into a paper-cutout sticker with a thick outline and a soft shadow, and builds a torn-paper collage world around you — while your face and clothing stay exactly as filmed.
What is the short version?
The whole look rests on a single contrast: everything in the frame is handmade paper except you, who stay photographic. Convert the person along with the rest and it stops being a collage and becomes an illustration, which is a completely different and far less arresting thing.
- Go to Vizard Agent and upload the clip.
- Give the palette — four colours is the right number.
- Say explicitly that your face and clothing must stay real.
What do you need before you start?
A short clip and a firm palette. Vizard Agent builds the collage world from generated textures, so the constraint that matters most is the colour list — an editorial collage with six colours in it stops looking like a designed object.
- The clip. Ten to fifteen seconds.
- The palette. Four hex values, including the paper.
- The preserved subject. Face, skin tone, clothing.
- The banned effects. Neon, emoji, futuristic gloss.
- The words to stamp. Which phrases get emphasis.
What do you type into Vizard Agent?
Spend most of the brief saying what must not happen to you. Vizard Agent will otherwise stylise the subject along with everything else in the frame, which is the single mistake that kills this look entirely, so the prohibitions are worth more here than any amount of description.
Prompt
Variants worth knowing:
- Four colours, with hex values. Including the paper stock.
- Ban the stylisation of the subject. Explicitly, by name.
- Ask for the stamps. Key words as printed cutouts.
What does Vizard Agent actually do?
Here is the order Vizard Agent worked in on a real collage piece cut from an eleven-second clip. Note steps eight to eleven, where it goes looking for the right filter in the ffmpeg build rather than assuming one is available.
- Checks the clip's duration and gets the transcript to match words to timings.
- Extracts frames to see what the speaker is actually talking about.
- Removes the background from the speaker to turn her into a cutout.
- Checks the motion-graphics and image tools available for the collage elements.
- Generates the background, torn paper textures and graphic elements.
- Creates the word stamps for the phrases being emphasised.
- Applies the sticker effect — charcoal outline, paper texture, soft shadow.
- Checks which morphology filters the installed ffmpeg actually has and reads their options.
- Tests the sticker effect on a single frame before committing to the full clip.
- Renders the speaker sticker for the whole duration.
- Layers the sticker, logo, stamps and badge over the paper background.
- Composites with snap zooms and a stop-motion jitter, synchronised to the audio.
- Reviews frames from the render and checks the torn-paper wipe transition.
- Re-renders several times to fix a black background caused by the filter graph.
- Simplifies the filter graph, confirms the cream base is visible, and delivers.
Steps thirteen and fourteen are the honest part. The background came back black more than once because of how the layers were being combined, and Vizard Agent kept looking at the frames and rebuilding the graph until the paper was actually there — rather than delivering a render it had not checked.
What does the result look like?
The finished piece this page is written from measured 720x1280 at 24fps and ran 10.01 seconds — a vertical clip with the speaker as a full-colour paper sticker, outlined in charcoal with a soft shadow, over a cream halftone paper ground with torn newspaper wipes and printed word stamps.
Her skin tone, clothing and face are untouched. Everything around her is paper. The motion carries a slight stop-motion jitter, which is what makes it read as handmade rather than as a filter.
When does this not work well?
The style is strong and opinionated, and it does not suit every subject or every message. Vizard Agent will build it faithfully to your brief, which means an unsuitable topic ends up with a well-made collage that is simply wrong for what it is trying to say.
- Serious subjects. The look is playful whether you want it or not.
- Poor background separation. The cutout edge gives it away.
- Long clips. Ten to fifteen seconds is the useful range.
- Too many colours. Four is a design; six is a mess.
- Complex hair or motion. Cutout edges struggle.
How do you fix a result that came back wrong?
Say which layer is wrong rather than describing the overall feel. Vizard Agent keeps the subject cutout, every generated texture, the word stamps and the filter graph itself, so a heavier outline or a different paper ground is a re-composite rather than a full rebuild.
- "The outline is too thin." Re-rendered at a heavier stroke.
- "My skin looks graded." Subject restored to the original colour.
- "The background is too busy." Textures thinned and re-layered.
How does Vizard Agent compare to doing it yourself?
By hand this is a motion designer in a compositing suite: rotoscoping the subject, sourcing paper textures, building the stroke and shadow, animating stamps. Vizard Agent generates the elements and composites them, then keeps looking at the frames until they are right.
| By hand | Vizard Agent | |
|---|---|---|
| The cutout | Rotoscoped or keyed | Background removed from the footage |
| Paper textures | Sourced or scanned | Generated to your palette |
| The sticker treatment | Built per layer | Outline, texture and shadow in one pass |
| Verification | Watch the export | Frames extracted and reviewed each render |
Common questions
Should the person be stylised too? No, and this is the whole trick. Vizard Agent keeps you photographic against a paper world.
How long should the clip be? Ten to fifteen seconds. Vizard Agent can go longer but the style tires.
Do I need a green screen? No. Vizard Agent removes the background from ordinary footage.
How many colours? Four, including the paper. Vizard Agent works to the hex values you give.
Can it add word stamps? Yes. Vizard Agent times them to the words from the transcript.
What is the jitter for? It reads as stop-motion. Vizard Agent adds a slight frame-level shake so it feels handmade.
Can it use my logo? Yes. Vizard Agent layers it in as another paper element.
Will the cutout edge look clean? Usually. Vizard Agent tests the sticker effect on one frame before rendering the clip.
Can I get other collage styles? Yes. Vizard Agent takes a different palette and set of textures for a different era.
Does the audio change? No. Vizard Agent keeps your original speech and syncs the animation to it.
Why does the palette matter so much? Because handcrafted editorial design is a discipline of restraint. Vizard Agent will use six colours if you list six, and the piece will stop looking designed.
Can it do this to an existing published video? Yes. Vizard Agent works from whatever footage you upload.
Does it check its own render? Yes, repeatedly. Vizard Agent extracts frames after every render and rebuilds when the layers are wrong.