Next-gen scan posters
A poster style built around the client's real imagery, re-shot as "scans": iso-line contours on a face, a thermal heat map, glowing wireframe edges, a LiDAR point cloud. It sits on a near-black 28 px grid with a single accent colour, condensed display type and small monospace HUD tags. It reads as engine / XR / digital-human tech without inventing any fake product shots.
What is in here:
| Path | What it is |
|---|---|
scripts/scan.py | photo → scan art (contour, thermal, edges, points, grade); always fits without stretching |
scripts/render.py | HTML → A4 300 dpi PNG + PDF + preview, plus an options montage; finds Chrome / Edge / Chromium |
scripts/demo.py | runs the whole pipeline on a procedural image; use it once to check the machine works |
assets/split-hero.html | half original / half scan hero; also the per-item page of a numbered series |
assets/portrait-callouts.html | one big scanned figure with callout marks, text column on the left |
assets/scan-grid.html | 3×3 grid, each item in a different scan mode, plus an info cell |
references/style-guide.md | tokens, type, layout numbers, scan-mode recipes, series theming, pitfalls. Read it before the first poster |
Needs Python 3 with Pillow and numpy, and a Chromium-based browser.
Workflow
-
Collect the content first. Logos, the client's own photos / renders, the headline, the real facts (dates, formats, price or offer, address), the QR code. Copy that describes a course or product should come from the client's own material (website, brochure). When nothing exists and you write a line yourself, tell the user which lines are yours.
-
Pick the layout per poster.
- A face or figure →
portrait-calloutsor a split. - One hero scene →
split-hero. - Several items of equal weight →
scan-grid. - One poster per item →
split-heroas a series, see below.
- A face or figure →
-
Make the art with
scan.py.- Size every image at 2× the CSS box it fills: hero 1588×1080, portrait 1240×1560, grid tile 700×584.
- Choose
--boxand--centerso the subject sits where the layout wants it. - For a split, build
real.jpg(modegrade) andscan.jpgwith identical--box/--size/--center, or the halves won't line up. - Colour the scan with the poster accent (
--color).
-
Copy a template next to an
art/folder and edit it.- Set
--A(accent),--B(secondary) and, for a split,--sx. - Replace every placeholder line.
- Put the real logo files and QR in place: an
<img>inside.qr, logo<img>/<svg>inside.top.
- Set
-
Render.
python scripts/render.py a.html b.html c.html --montage options.jpg. -
Look at the previews yourself before showing anyone. The usual faults:
- a stretched image
- the split off the subject
- a callout on hair instead of the face
- a headline running into the big number
- a one-character orphan line
- unreadable small tags over bright art
Fix them and re-render.
-
Show options, then package.
- When asked for a design, make three distinct versions and send the previews or montage.
- Produce final files only for the one the user picks.
- Name output files in ASCII.
Series (one poster per course / product)
Use split-hero.html for every item and keep the layout identical so the set reads as one family. Per item, change four things:
- accent colour: the item's own brand colour if it has one, applied to the scan tint, number, tags and offer
- hero image and scan mode: vary the mode across the set (points, edges, thermal, contour…)
- the series bar: highlight this item's cell
- the big outlined number
Keep each item's explanation to one sentence plus three short points, taken from the client's own course or product copy. Shrink the title for long names so it clears the number, for example 54 px instead of 64 px.
Things that matter (and why)
- Never stretch imagery. Crop, then fit (
scan.pydoes this). People notice squashed faces immediately; it was the first complaint on every early draft. - Put the split line through the subject's centre, between the eyes and down the nose for a face. Off-centre, it reads as a layout accident rather than a "before / after scan".
- Callouts point at real features (face, hand, fabric). Check them in a render and move the SVG marks; a mark on the hair or the background looks careless.
- Decorative labels must not read as claims. "SCAN · THERMAL", "MODE · POINT CLOUD" and legend words are styling. Keep numbers out of them, and don't imply the product does something it doesn't.
- Effects never cost legibility. For a glitch headline, keep a clean, fully opaque copy of the text on top and put the RGB-split / sliced copies behind or in thin bands. See the style guide.
- Small text over art gets a backing. Use a semi-opaque dark chip or a text-shadow on the ground colour.
- Fonts. The templates pull Barlow Condensed, JetBrains Mono and Noto Sans SC from Google Fonts, with fallbacks to Bahnschrift / Impact, Consolas and Microsoft YaHei / PingFang. Offline or in mainland China, the fallbacks are what renders, so check the preview. For guaranteed output, install the fonts locally.