Your brand's look now reaches everything on its own. **Visual analysis** is switched on by default both when you add a brand and when you re-analyse one, and it now reads your store's stylesheets as well — so it finds your actual colours (with exact colour codes), your typeface and your shapes, not just the handful of colours visible in the homepage markup. The **Visual style** field became short and factual as a result: three lines that every module reads the same way. Image generation, newsletter templates and Shopgrade's styled product descriptions all use it (and your AI influencers get the same three lines) — and for the latter you **no longer need to paste colour codes into your prompt**, it uses your store's colours by itself (a colour named in the prompt still wins). If a re-analysis would overwrite the style you fine-tuned by hand, we warn you first.
Your monthly Ninja AI budget now stretches much further: a ten-question conversation costs a fraction of what it used to, while the answers stay the same — in fact it remembers the recent part of the conversation more precisely. And if the budget does run out, the chat no longer goes silent: you can still ask questions and it can still look at your data, it just won't perform actions (creating posts, editing, publishing) — instead it shows you where to do it yourself in the app. We warn you at 80% of the budget, and when a conversation gets long, the chat offers to start a new one for your next topic. [A Ninja AI havi kerete](/help/ninja-ai-havi-keret)
The free product page diagnosis now starts with your email address: you tell us where to send the result, and the evaluation runs right after. If you arrived from our website's form, the field is already filled in — just hit Next. In return your result no longer gets lost: if you gave us your address but didn't run the diagnosis, you get a reminder the next day that picks up where you left off; if it ran but you didn't ask for the coupon code, you get a message with your own score and why your weakest point is a problem for you, and two days later one about what happens if you leave it as it is. Three messages at most, and as soon as you run the diagnosis or ask for the coupon code, the remaining one is dropped automatically. Every message has an unsubscribe link at the bottom. A shared result page still opens without any email address.
Fixed
Content is no longer created while we silently skip something. If we can't work with a setting you switched on — no knowledge source selected, no link on the post, or no image tool chosen — we now ask before **Generate**: **Beállítom** (I'll set it up) or **Mégis generálom** (Generate anyway). No credit is spent until you decide. If an automation or Ninja AI created the content, the warning stays at the top of the post (with a ⚠ marker in the list), and the run log shows a **Részleges** (Partial) row with the reason. Image generation from a product link got better too: we now also use the link written **in the post's text** (until now only links saved during generation), so posts brought over from the old system or written by hand get your actual product on the image instead of an invented one — and if the link yields no image, we tell you that as well. If a generation gets stuck, after three minutes we say it didn't start so you can try again. [Hiányos generálás: mit jelent a figyelmeztetés](/help/hianyos-generalas-figyelmeztetes)
A webshop knowledge source brought over from the old system now really comes to life the moment you connect your store. Until now the migration carried over every setting of the source but — for security reasons — not your store's key, and nothing tied the two together: the source silently returned nothing, and your automations produced generic posts instead of ones about your products. From now on, saving the connection links your waiting sources automatically (the confirmation even tells you how many), and it works the other way round too, if the store was connected first. If you have several stores we don't guess for you: the source visibly stays "Waiting to be connected" and you pick its connection. On top of that, a misconfigured source is no longer silent: it gets a badge in the Knowledge sources list, and a red warning next to its name when generating, telling you what's wrong and where to fix it.
The Publish step of a post now only lists the accounts we can actually publish a post to: your Facebook page and your Instagram. Until now your newsletter and blog accounts showed up there too, marked "Not connected" — and if you ticked one, publishing got stuck on it. You now see the same list here as on the Text step. Newsletters are still written under Newsletters, blog articles under Blog articles.
