What's new in SpikeReel: framing that follows the cut, captions that get the words right, and a headline you place yourself

Most of this week's work is invisible in the interface and obvious in the clip. We took real projects from real creators, looked at the frames and listened to the captions, and fixed what a viewer would notice. Here is what changed.

Auto framing decides per shot, not per clip

A produced show cuts every few seconds: the host to camera, a two-shot at the desk, a phone screenshot filling the frame, a graphic. One crop for the whole clip cannot serve all of those. Auto is now the default framing and it decides shot by shot: it crops to the face while there is one, and shows the whole frame on a blurred field the moment a shot has nobody in it, then crops again on the next cut. Extreme close-ups are shown whole too, so a documentary insert no longer becomes an eye filling the screen. Fill and Fit are still there when you want to force one look.

The headline stays off the face, and you can place it

The opener headline used to be placed from a model of where the face would land, and every gap in that model put a title across someone's eyes or in a different spot on every clip. It is now placed by looking at the actual opening frames of the actual output, so it lands the same way every time and stays clear of the speaker. And in your brand template you can now choose where it goes: keep it clear of the face automatically, or pin it to the top, the centre, or the bottom just above the captions.

One more thing you will notice: the cover and the first frame now carry one and the same headline. The poster reuses the exact title the video burns in, so nothing changes the instant a clip starts playing.

Captions: a second listener for every language

Speech recognition is very good and still mishears: a dialect word becomes its formal look-alike, a name becomes a common noun, two words fuse. Every transcript now gets two things. First, the recognizer is told the video's own title and description before it listens, so it knows the names and the register it is about to hear. Second, a language model reads the whole passage and corrects misheard words, one word at a time, without touching timing or dialect. On an Egyptian Arabic episode that fixed 136 words; on English football commentary, 17. It runs for every language.

Your brand, in every language

• Fonts are chosen per language. Pick a face for Arabic, another for English, another for Japanese, in one template; each clip gets the face of the language it is actually spoken in. Ten scripts are covered, from Cyrillic to Thai.
• Caption and headline presets preview in the language you are designing for, so you see how Arabic or Korean will really look before you render.
• The end card now wears each platform's own look, and its spoken call to action is written and voiced in the clip's language, 23 languages so far.
• Arabic titles and project names now read correctly in the studio in every interface language.

Smaller things that add up

• More sources: paste links from Twitch, Dailymotion, Loom, Zoom recordings, Google Drive, Dropbox or a direct video file, not only YouTube.
• Team seats: Pro includes one extra teammate, Creator three, working in the same library.
• The calendar follows your own timezone, and nothing can be scheduled into the past any more.
• Spike Agent has a new face and a launchpad that shows what it can do, and it follows a running job live.
• If a video is too short for the number of clips you asked for, the project now says so instead of quietly delivering fewer.