Your videos and narrator videos can now be about your real product — no more pasting its details in by hand. In the Vizuál step, above the Utasítás field, both the 🎬 Videógenerálás and the 🎙️ Narrátorvideó block gained a Tudásforrás hozzáadása switch: turn it on and the AI also draws on the text of the selected item — product description, features, price — picked Kézi, AI választ or Véletlenszerű the same way you know from text generation. This setting supplies the text; the knowledge source under Bemeneti kép supplies the image — so for a video about one specific product, pick the same item in both places in Kézi mode. On videos it needs prompt optimalizálás switched on; if it isn't, a quiet line under the switch tells you why. Narrator videos also gained an Anti-AI mód: with it on, the AI writes a more natural, human-sounding narration and avoids machine-like phrasing (_„a mai rohanó világban”_ and the like). Both settings are saved into templates, and if the source happens to be unreachable the video is still produced without it — and the post tells you what was left out.Help:Videó és narrátorvideó generálása
You can now get a full article brief for your own keywords — no need to wait for the machine to find them. The 3. Egyéb beállítások block of the usual research form gained a Saját kulcsszavak field: you type your exact phrases there, one per line (up to 25), and the full brief is built for exactly those — search volumes, Google results, suggested title, outline, questions and internal links. One round can mix machine search and your own keywords — or, with the quantity picker's new „ne keressen, csak a sajátjaimat” option, only yours. The credit estimate in the footer splits the two before you start (1 credit per keyword, the same as for machine-found ones). Your phrase is never filtered out for low search volume or relevance, and if Google has no data for it the row still goes in — the numbers are replaced by _nincs keresési adat_. Anything already in your set or previously rejected is called out up front, and no credit is charged for it: a Visszaállítás (ingyen) button puts a rejected keyword straight back with its existing brief. Your own rows carry a quiet kézi badge.Help:Saját kulcsszó felvétele
Changed
Images no longer re-download on every page change — and the Galéria finally stops pulling megabytes it doesn't need. Until now the same image was downloaded again on every list refresh and page change (the browser couldn't tell it already had it), and the Galéria grid loaded full-size images just to show them at thumbnail size. Image addresses are now stable while they stay valid, and the gallery loads thumbnails: a 48-image page is 2.3 MB instead of 29 MB — most noticeable on mobile. Click an image and the editor still shows it full size. The same applies to the calendar, the post and blog lists, newsletter logos, saved prompt covers and profile pictures.
Your keyword set is now available as a knowledge source in the blog writer on its own. Until now the set only appeared among knowledge sources once you had clicked Cikk írása at least once — until then your `/knowledge` list was empty and you could not set up a keyword-based automation either. Now it is created automatically at the end of the first successful research round, named after the research; renaming the research renames the source too (unless you gave it your own name). If it currently holds no approved, unused keyword, a red notice in the source picker tells you what to do. The research header also states that the set can be used as a knowledge source.Help:Cikk írása egy kulcsszóból
The keyword list now uses the screen, and no longer scrolls away from under you. On the research page the keyword list on the left used to stop at 65% of the screen height, so with a 25-item set you could not even see half of it — while the space next to the article brief on the right sat empty. Now the list grows to the height of the screen and stays put while you read the long brief: the filter icons and the Összes kijelölése row stay visible while scrolling, and switching to another keyword is one click. Nothing changes on a phone: the page scrolls there as before, and the list runs its full length.
At the end of a research round we now tell you what came of the idea you typed into Saját téma. Until now the topic you gave could be swallowed silently — it went into the starting words, but the suggestions it produced could be dropped by the filters, and nothing said so. The round summary now states how many keywords came from it; if none did, we say that too, and point you at the Saját kulcsszavak field.Help:Kulcsszókutatás indítása
Keyword research now takes your plain website address — you don't need to know where your sitemap lives. In Knowledge sources you can now type `mysite.com` into the Oldaltérkép (Sitemap) field: Teszt finds the sitemaps, writes the addresses it found into the field, and tells you how many there are — delete the ones you don't need (a foreign-language variant, say) before you save. If there is no sitemap at that address we say so plainly and leave the field untouched. The research source field is now called „Melyik boltból vagy sitemap-ből?” — the label itself says you don't need a connected store — and if you do have a sitemap source but it belongs to a different brand, the dropdown says so instead of opening empty.Help:Sitemap tudásforrás beállítása·Kulcsszókutatás indítása
Background music for narrator videos is temporarily unavailable — and we now say so. The provider that supplied the music withdrew the service, but the switch could still be turned on: you got a silent video, the system logged it as a success, and the credit was charged. The Háttérzene switch is now greyed out, its name says _átmenetileg nem elérhető_, and the `?` explains what you get instead. If an old template still starts a video with music turned on, the video is produced — narration and captions included — and the post shows that it was finished without music. No credit is charged for music.Help:Videó és narrátorvideó generálása
Fixed
You can browse your finished images and videos while a generation is running. While a generation was in progress the Vizuál step refreshed every 3 seconds, and each refresh made the browser re-download every image and video: the large view flashed (it looked as if you could not switch between images), a video you had started jumped back to the beginning roughly every 5 seconds, and your selection slid over to the neighbouring image whenever a new item finished. Now the image stays put, the video plays through, and the selected item stays where you left it — a new item lands at the front of the strip without stealing focus. The page also downloads far less data. The same applies to the public sample post. Separately: a full-size image or video now closes when you click beside it too — with a portrait video a ~275-pixel band next to it used to swallow the click, so in practice only `Esc` closed it.
Your Facebook connection is no longer flagged red as "Leállt" just because Facebook withholds one piece of data. The nightly refresh also asked for the per-post statistics of your page's entries — something Facebook does not grant our app access to — and we treated that refusal as an expired connection. That put a "Leállt — kösd újra" badge on 32 perfectly working connections that post every day, and re-connecting fixed nothing, because there was nothing to fix. Only genuinely expired access counts as a breakdown now. And in Analitika we say out loud what used to be misleading: no Facebook post can appear in the Top posztok list, because Facebook does not give us per-post numbers — your account's aggregate figures (followers, reach, engagement) are there as always, and the old _"a következő frissítés pótolja"_ message no longer shows up for this reason.Help:Analitika: hogyan teljesítenek a posztok
The public sample-post page loads again. Since September 4 every visitor got a blank error page on the very page our ads and website send people to so they can try the product. Fixed — along with the thing that let it happen quietly: the next error of this kind breaks on our side, not on the visitor's.
Publishing error messages no longer send you the wrong way. When Facebook or Instagram rejected a post we sometimes told you to _check the post's content_ — even though nothing was wrong with the post and a video upload had simply been cut short. You now get the real reason and what to do about it; and when we don't recognise the reason either, we say so honestly instead of sending you on a pointless review. Messages that stated the same thing twice are gone as well.
Image generation no longer fails on a single model hiccup. If the model answers with text instead of an image, we now retry once before telling you — so you get an image instead of an error more often, at the cost of a few extra seconds.
If a generation is cut short, the system picks it up on its own — you no longer have to restart it. Until now, if a job died mid-run (for example because we were deploying an update), the "in progress" tile eventually turned into an error and you had to start over; and if you had closed the tab, nothing happened at all. A watchdog now checks for stuck jobs every few minutes and restarts them automatically — the tile keeps saying "in progress", because the image really is coming. It costs no credit: we only charge for the image that is actually produced. If it fails twice, you get an error and a retry button. This works for post and blog text, images, animations and narrator videos.
