Posting style learned from your accounts

Where: Sparky AI > Knowledge, the "Posting style, learned from your accounts" panel.

Ignix looks at what your connected accounts have already published and works out three things, so that anything it creates for you looks and sounds like you instead of like a stock brand:

Your colours, each with its role: which one is the accent that identifies the brand, which is the background, which is the text colour, and the ends of any gradient or glow.
How your posts are built: layout, typography, treatment, the elements that repeat across your posts (logo placement, frames, grids), and what would look wrong for your brand.
How you write, per network. A brand does not write the same way on X as on LinkedIn, so a network with at least four published posts gets its own description; the rest fall back to a general one.

This shapes NEW POSTS ONLY — the composer and Autopilot. It does not change how Sparky replies in chat, DMs or comments; that is set in the Knowledge sections of the same page.

What it learns from

Only posts that actually reached a network, including the ones published natively outside Ignix. Images you generated with AI but never published are ignored on purpose, so the AI never learns from its own output.
Images only. From a video post it uses the thumbnail, which is a real frame of yours.

When it runs

Automatically once a week. A brand's style does not change from one day to the next.
Or immediately, with the "Learn now" button, which is what to use right after connecting your accounts for the first time.

If the panel shows nothing, you have no published posts with images yet. If it shows your colours and how you write but no "How your posts are built" block, an amber note explains exactly why the images could not be read.

The captured voice is a starting point, not a verdict: you can edit the text of any voice block by hand, and once you do, the weekly sweep never overwrites it ("Back to what was learned" undoes your edit). Colours and composition are measured, not written: if a colour looks wrong, press "Learn now" again after publishing posts that better represent the brand.

Style per account

Below the workspace panel sits "Posting style, per account": one collapsed card per connected account. Each account learns its OWN palette, its own composition guide and its own voice, from what that account publishes — three visually different Instagrams keep three different styles, and content generated for an account uses what THAT account learned. The workspace style is the fallback while an account has not learned yet.

An account's voice is learned from all of its posts, with or without images — an X account that only posts text still learns how it writes. The palette and the composition guide need posts with images.

Style examples you add

Where: inside each style card, the "Style examples you add" block.

Have posts on accounts that are NOT connected to Ignix — an old TikTok, another brand of yours, a client's account you run elsewhere? Paste their captions or upload their images there and the learner mixes them with what your connected account publishes. It is how a brand-new account can already sound like one you have been cultivating for years. Nothing you add here is ever published: it is only used to learn.

You choose where it goes. Open the block inside the general card and it feeds the workspace style; open it inside an account's card and it feeds only that account.

What to add, and how much

TEXTS: one post per block, separated by a blank line (or a line with ---). Around 10 useful captions is the minimum, 20 is plenty and past 50 it stops improving. Variety matters more than volume: ten different posts teach more than ten identical promos. Anything under 15 characters is dropped, and so are duplicates.
IMAGES: 6 to 12 pictures from real posts. Pictures, NOT screenshots of the post — a screenshot brings in the app's interface (its whites, its blues, its font) and poisons the measured palette.
The minimums are a warning, never a wall: with less material it still learns, and the counter tells you where you stand.

VIDEO ACCOUNTS. This is the answer to a real gap. From YouTube and TikTok we read titles and descriptions, never the video itself — no transcripts and no frames — so what is learned is a headline voice, not the way you actually talk. Paste the transcripts or scripts of your videos and that gap closes: a transcript is the richest voice sample there is. YouTube generates automatic subtitles you can copy in two clicks, and thumbnails uploaded as images give the palette.

Removing it

Each example has its own bin, and there is a "Delete all" for the whole destination. The automatic re-learning never deletes what you added — it is material you put there on purpose — but removing it is always one click away. After deleting, press "Learn now" in the same block so the style goes back to what your connected account shows, instead of waiting for the weekly sweep.