Make it look good

Good video and bad video usually contain the same ingredients. The difference is almost never the effects list. It’s whether the visual decisions were made once, upfront, as a system — or invented one layer at a time.

Improvised design drifts. Three nearly-identical greens. Headline sizes that wobble between scenes. A purple gradient nobody asked for. Everything animating at once. None of it is a bug, and all of it reads as amateur.

This page is the antidote, and it works the same whether you’re building by hand or briefing an agent.

1. Decide your style before you compose

Before the first layer, settle a short list of values and then treat it as the only source of visual decisions for the whole piece:

• Colours by role, not by vibe. A background, a surface, one text colour, one accent. Two or three from the same family beat a rainbow every time.
• A type pairing — usually one family with a real weight jump, or two families with clearly different jobs. Decide the headline size, the body size, and the gap between them.
• A spacing step — one number you keep multiplying (8, 16, 24, 48) rather than eyeballing each margin.
• One easing curve that the piece’s motion mostly uses, so movement feels like it comes from one hand.

Write them down. Then every value you use comes from that list. If a colour isn’t on it, it doesn’t go on the canvas.

Where the values come from

If you have a brand, they come from the brand — the real typeface, the real hex codes, the real logo, as assets. If the piece references something real, go and get its actual material. Invented lookalikes are the single biggest quality drop.

And be honest about the ceiling: your assets decide how good it can look. An asset-poor project needs material found, shot, or generated early — not worked around later with gradients.

2. Check real frames, not the layout

The canvas tells you what you asked for. Only a rendered frame tells you what a viewer gets.

Overlapping text, a headline that loses contrast over a photo, an entrance caught mid-flight, a card clipped by the frame edge — none of these are visible while you’re arranging layers. Scrub to a resting moment in each section and actually look at it.

Check as you go, not once at the end. A foundation problem caught in the first section costs one section; caught at the end it costs the piece.

What to look for

Judge each frame as someone seeing it for one second:

1. One focal point. If your eye bounces, the frame is overloaded — give something its own moment.
2. Readability. Every line legible at a glance, with real contrast against what’s behind it. Text over imagery needs a scrim.
3. Integrity. Nothing clipped, overlapping, hugging a frame edge, or caught awkwardly mid-animation. Elements either clearly group or clearly separate — near-touching reads as a bug.
4. Sized to the frame. Anything meant to reach an edge actually reaches it. Unintended bars or margins mean something wasn’t sized to the composition.
5. Designed, not defaulted. Does it look intentional, or does it look like the first thing that appeared?
6. Coherence. Palette, spacing and alignment consistent across scenes. One section in a different font breaks the whole piece’s voice.
7. No dead air. No moment shows bare background with nothing meaningful in frame. If content arrives late, tie its entrance to the moment rather than fading it in afterwards.

Check detail at the scale its defects show

Reviewing the whole composition proves the composition. It doesn’t prove the things inside it. Anything small, tilted, blurred or moving hides its own flaws — mis-centred text inside a narrow bar is invisible at 40% of frame and unmissable at full size.

So build detailed pieces — a device screen, a card, a chart — in their own composition first, flat and full size with no camera or filters. Fix them there, then place them in the real piece. A project holds as many compositions as you like, so a scratch one costs nothing — and it doubles as a component once it’s right.

3. Know the tells

These are the things that make a video read as generated rather than designed. Every one is invisible while you’re building and obvious on screen.

1. The stock palette — the purple-gradient family and its cousins, or any palette that ignores the actual brand.
2. Identical scenes with swapped colours — one layout run N times reads as a slide deck, not a designed piece.
3. Centred everything. Editorial asymmetry — a 70/30 split, an offset column — creates direction. Use at most one centred moment.
4. Polite type hierarchy — a 40px headline above 36px body. If the hierarchy isn’t unmissable, it isn’t hierarchy.
5. Low-opacity hedging — a motif at 5% opacity is a gesture at design without a decision. Commit or cut it.
6. Decoration that wouldn’t be missed. If deleting an element changes nothing, it was noise.
7. A rule or underline beneath every heading. The classic filler move.
8. A gradient where a photo should be. That’s a placeholder, not a design.
9. Default type everywhere — one font, regular weight, no tracking decisions.
10. Metronomic motion — everything moving at once, identical durations, linear easing, maximum energy from the first frame. Energy should build. A related version: everything enters by fade or slide-up and the big moment is a scale-up. Your piece’s signature move deserves more than a preset; presets are for the supporting cast.
11. Interface furniture in a non-interface piece — CTA pills, nav bars and search fields in a title sequence or brand film. Nothing on screen is clickable. A button belongs only when the frame is genuinely depicting a product. Otherwise your closing line is type, not a rounded rectangle.
12. One fat drop shadow — a single big soft shadow instead of a layered, background-tinted one. Banded gradients with no grain belong here too.
13. Numbered index labels everywhere — “01 /”, “02 —” stamped on every scene. It earns its place only when the numbers mean something the viewer needs — ordered steps, chapters, a ranked list — and then the piece commits to the full set.

4. Know when you’re done

The finish line is not “I rendered it.” It’s this:

A fresh look at the frames produces no concrete improvement you’d make.

If you look at your video and can name three things you’d change, you’re not done — you’re finished building. Naming problems and shipping anyway is the most common way a piece stays mediocre. Every issue you can name should end as an edit, or as an explicit decision to keep it.

One or two rounds of fixes is normal. If a frame still doesn’t work after three, the layout needs rethinking rather than nudging.

Agents are held to all of this

A connected agent loads this same doctrine as enforced skills: it derives a style foundation before composing, renders real frames and critiques them section by section, and audits its own work against the tells above.

Which means the useful thing to say to an agent isn’t “make it look nicer.” It’s to give it what only you have:

• “Use our real brand — here’s the site, pull the actual fonts, logo and colours.”
• “The hierarchy is too polite in scene two — make the headline unmissable.”
• “Scene three is centred like the others; break the symmetry.”
• “What would you improve about this frame?” — and then expect the fixes applied, not listed.