Not a translated interface. Full Hebrew in prompt writing, narration, captions and support, and where Hebrew genuinely hits a ceiling, this page says where it is and what to do instead.
The engines understand Hebrew well at the level of intent, and you can describe a whole scene in Hebrew and get exactly that. What is true is that professional cinematography terms work more precisely in English, simply because the engine was trained on English descriptions. The approach that works: write the scene in Hebrew and leave a term like shallow depth of field in English mid-sentence. That is not a compromise, it is what produces the best result.
An image model does not write text, it draws it as a shape it has seen before. Most of them saw billions of images with English text and very few with Hebrew, and learned a left-to-right writing direction, so Hebrew comes out reversed or as gibberish. The system has two engines that do write Hebrew correctly inside the frame, and those are the ones to route to when the wording has to be part of the picture. In every other case, generate a clean image with no words at all and add the wording as a separate layer, where you also get full control over the typeface and size.
Video engines that generate sound together with the picture cannot speak Hebrew, and the result sounds like accented gibberish. So Hebrew speech is not generated inside the shot but built separately: generate Hebrew narration in a natural voice, and if a character has to talk on screen, lip-sync them to that voice. It is also the only route that lets you fix a line after the shot exists, without regenerating it.
Hebrew is not a "yes" or "no", it depends on the field. Chat and writing: the big models write and understand Hebrew very well, for posts, scripts and replies to customers. Narration: only some voice engines speak natural Hebrew, the rest sound like accented English. Captions: Hebrew transcription is accurate on clear speech, and the challenge is the right-to-left direction. Images: most engines understand a Hebrew description, but only a few write Hebrew inside the picture. Video: a Hebrew prompt works, Hebrew speech is added separately. Whoever joins all these fields into one workflow skips most of the trial and error.
A business owner does not need a tool, they need a finished post. The short route: write in Hebrew what you want to tell customers, generate an image or a short video that shows it, add Hebrew narration or a Hebrew headline inside the image, and captions for viewing without sound. And anyone already working inside ChatGPT or Claude can connect Cadabra to them and ask for all of it in conversation, in Hebrew.
There are three ways to get Hebrew speech. Narration: write text and get a natural Hebrew voice, about 1 credit per 750 characters, with emotion directions inside the text. Your own voice: clone it once and it reads any text. And a talking avatar: one photo of a face becomes a character that speaks Hebrew to camera, with synced lips and a subtitle file with every video.
The fastest sign of bad Hebrew AI content is translated Hebrew: long sentences, stiff connectors and words nobody says. Three rules help: write the request the way you speak, ask for short sentences, and read it aloud before you post. Songs add one more layer: vowel marks on names and risky words, so the engine sings them right instead of guessing.
Narration engines that support Hebrew read any text in a natural voice, and you can also clone your own voice. A talking avatar turns one photo into a character that speaks Hebrew to camera. Video engines that make sound inside the shot still do not speak Hebrew, so the voice is added as a separate step.
No. The interface, the explanations, the support and the ready-made prompts are all in Hebrew, and you can write a prompt in Hebrew and get a result. English only helps with professional cinematography terms, and even there it is a word or two inside a Hebrew sentence.
Because the model draws letters as shapes and learned a left-to-right writing direction from hundreds of millions of examples. No prompt forces it otherwise. Two engines in the system do solve it, and in every other case the wording is added as a separate layer.
Yes, because they are transcribed from the finished film rather than from the script. Transcribing from the script looks fine in a list and comes out drifting on screen, because what was actually said is never identical to what was written.
Yes, as are all the explanations and guides inside the system.
There is no single best at everything. For writing, the big models; for narration, a voice engine that supports Hebrew; for text inside an image, one of the two engines that write Hebrew correctly. In Cadabra they are all on one screen, and Hebrew is routed automatically to an engine that knows it.
Hebrew captions cost no credits at all. For everything else you buy a credit pack, and the price of every creation is shown before it runs.
Home · Pricing · What's new · All engines · All effects · Help center · Lessons · Meital Sela · AI video · AI images · AI voice · AI music · Image to video · What it costs · Credits, not a subscription · Blog · Connect to Claude and ChatGPT · Presets · Privacy · Terms
Cadabra AI · Meital Sela · Rehovot, Israel