LinkedIn carousels and post visuals
LinkedIn's document carousel is the highest-reach organic format on the
platform, and it has one detail everyone gets wrong: the bottom of every slide
is covered by LinkedIn's own UI. Canvora's LinkedIn formats declare safe zones
for exactly this. Setup and credits: see the canvora skill.
Format choice
| Format | Ratio | Bottom safe zone | Use |
|---|---|---|---|
linkedin_carousel_portrait |
4:5 | 60px | the default — most feed space per swipe |
linkedin_carousel |
1:1 | 60px | when their existing template is square |
linkedin_carousel_vertical |
9:16 | 80px | mobile-first, maximum height |
linkedin_post |
16:9 1200×627 | — | single-image post |
linkedin_article |
16:9 1920×1080 | — | article header |
Default to linkedin_carousel_portrait unless the user says otherwise. All
three carousels are 15 credits per slide. State that cost in the same message
as the run (a 7-slide carousel = 105) — announce it, don't ask permission. If
you chose the slide count rather than the user, confirm that number with them.
Intake
- What's the point of view? LinkedIn rewards a stance, not a summary. Get one sentence of opinion, not a topic.
- How many slides? 6-8 is the sweet spot; match any promised count. Max 10.
- Personal brand or company brand? Company →
--brand+--logo. Personal/founder posts often read better without a logo on every slide. - Is there a CTA? Newsletter, demo, comment-bait →
--cta.
The canonical run
ID=$(canvora generate \
--idea "Most B2B pipelines die at handoff, not at close — 6 fixes for the sales-to-CS transition" \
--format linkedin_carousel_portrait --slides linkedin_carousel_portrait=7 \
--brand $BRAND --style minimal --cta | head -1)
canvora status "$ID" --json # repeat every ~30s until completed
From content the user already wrote (a post, a newsletter section):
ID=$(canvora generate --input "$POST_TEXT" \
--format linkedin_carousel_portrait --slides linkedin_carousel_portrait=6 \
--brand $BRAND | head -1)
canvora status "$ID" --json
Both are 3+ outputs, so fire-and-poll rather than --wait — most harnesses
kill a shell command at 120s while the generation carries on server-side. Never
re-run generate after a timeout; that starts a second, separate job. Poll
status. See the canvora skill's "Long runs" section.
Content shaping for LinkedIn specifically
- Slide 1 is a claim, not a title. "Your pipeline is leaking at handoff" beats "Sales to CS handoff". The cover decides whether anyone swipes.
- One idea per slide, in professional register. No hype, no emoji walls.
--style minimalorcorporatematches the room;boldonly for contrarian/opinion posts. - Keep text off the bottom edge. The pipeline handles the safe zone, but don't request bottom-anchored text — LinkedIn's UI will cover it.
- Last slide earns the follow.
--cta. - Don't art-direct slide by slide. Do carry a stated brand look through.
After generation
- Sort by
sortOrder, then upload as a document post (PDF) — that is what LinkedIn treats as a carousel. MCP clients can produce the PDF directly:export_visuals(generation_id, format: "pdf"). CLI users download the images and export from the Canvora dashboard. - Also write the post caption — the visuals don't carry the algorithm, the first two lines of the caption do. That's your job, not Canvora's.
- Fix one slide with
edit(3 credits), never a full regeneration.
Quality checklist
- Portrait 4:5 unless the user's template says otherwise
- Cover states a claim; 6-8 slides; count matches any promise
- Professional register;
minimal/corporateunless the take is contrarian -
--ctawhen there's an ask - Delivered as a PDF document post, slides in
sortOrder - Caption written alongside the visuals