I03 — Four press photos, a carousel ready to post

The photos of an event arrive in the evening; the carousel should be out before the morning. It needs a cover that stops the thumb, slides that look like one story and not four photographers, and words: a caption per slide, a text for the post, hashtags, and — for everyone who reads with a screen reader — an alt text per picture. In two languages, for this account.

This example does it in one run from four photos and two lines. The model draws a cover with its title in it; the deterministic steps give every slide the same look, size, number and mark; a vision model looks at the photos and writes the words, as data the publishing tool can read.

The exact graph is the workflow (as a diagram).

The case

A science account posts a recap of a night launch from NASA’s press photos — on the pad, lift-off, the reflection in the water, the ascent — credited to NASA.

Input	Who provides it	Reference run
Photo 1–4	the photo desk, in story order	NASA press photos of the Artemis I launch
Event	the editor	Artemis I lifts off
Where and When	the editor	Kennedy Space Center · 16 November 2022
Account Mark	the account	its orbit mark, an SVG

What it shows

• A cover with words in it. A Text Template writes the brief from the two lines; Image Transform on GPT Image 2 redraws the last photo as a vintage travel poster with the title and the line in it, spelled as given and kept clear of the edges; Topaz doubles it. The cover is then placed whole on the slide — a finished page is never cropped and never gets a badge over its words.
• One look for four photographers. The same Curves on every photo, the same 1080 × 1350 frame, a number top right and the mark bottom left: four photos read as one story (Consistency and brand).
• Words that come from the pictures. Image Analysis reads the four photos in one question and returns JSON: per slide an alt text (at most 160 characters) and a caption in English and Italian, then the post in both languages and eight hashtags. The question says to describe only what is visible and not to invent people, names or numbers.
• Almost all of it free. Ten credits buy the cover; everything else — look, sizes, numbers, marks, the sheet — costs nothing, and the words a few hundredths of a credit.

The workflow, step by step

Phase	Steps	What they do here
A — cover	A1 Write the Cover Brief, A2 Poster Cover, Text by the Model, A3 Cover Upscaled × 2	the brief from Event and Where and When; GPT Image 2 Edit, quality: medium, 3:4; Topaz × 2 in illustration mode
B — slides	B0 Mark at 90 px; B1 Cover on 1080 × 1350, Never Cropped; per photo Bka House Look, Bkb 1080 × 1350, Bkc Number k/5, Bkd Sign Slide k	the cover placed whole on a cream slide; each photo graded, cut to 4:5, numbered with Text on Image, signed with Overlay
C — words	C1 Write the Question, C2 Captions, Alt Texts and Post	the question from the two lines; Image Analysis with the four photos, response_format: json_object
D — approval	D1 Contact Sheet	a Grid of the five slides

The reference run

The words, as delivered (the JSON), begin:

{
  "slides": [
    {
      "alt_text_en": "A tall white-and-orange rocket begins lifting from a launch tower at night, surrounded by bright flames and billowing smoke.",
      "caption_en": "Fire blooms beneath the white-and-orange rocket as it rises from the tower into the night.",
      "caption_it": "Il fuoco fiorisce sotto il razzo bianco e arancione mentre si solleva dalla torre nella notte."
    },
    …
  ],
  "post_en": "A night launch turns the pad into a sea of fire. Smoke billows outward as the rocket climbs. Photos: NASA",
  "post_it": "Un lancio notturno trasforma la rampa in un mare di fuoco. Il fumo si espande mentre il razzo sale. Photos: NASA",
  "hashtags": ["#ArtemisI", "#NASA", "#RocketLaunch", "#Spaceflight", "#MoonMission", "#NightLaunch", "#SpacePhotography", "#KennedySpaceCenter"]
}

Output	Result
slide_1	the cover, 1080 × 1350 — full size
slide_2 … slide_5	the four photos, 1080 × 1350, numbered and signed
captions	JSON: four slides with alt text and captions in two languages, the post in two languages, eight hashtags
contact_sheet	the five slides side by side, for approval

10.1 credits: 5 for the cover, 5 for Topaz, 0.08 for the words. The words change a little at every run; read them before posting, as you would a colleague’s.

Adapt it

• Another event. The workflow knows nothing about rockets: a match, a concert, a product launch, a graduation — change the photos and the two lines.
• Another look. The cover’s style is a sentence of A1; the house look is the Curves of Bka, or a LUT of your own.
• More slides. Add a photo input and a branch of four B steps, and one more picture to C2; or move to a template when the count varies (Layouts, batches and codes).
• Other languages. Ask C2 for them; the vision model writes many.

Install it

node madoo-install-example.mjs i03-event-carousel --api <your Madoo API URL> --key <key.json> --run --save carousel

Photos: NASA/Joel Kowsky, Artemis I launch, Kennedy Space Center, 16 November 2022 (NASA images, public domain). The account and its mark are invented for this example and are not affiliated with NASA. The cover and the words were made by the models named.