Genien
Start free

What's new

What changed, and what's next.

Genien is in early beta and changes every week. This is what changed, with the real dates, and what we plan to do next. Plans can change.

Next

  • Online checkout, so that paid plans no longer need an email.
  • A direct Meta connection, once Meta's app review allows it. Until then you download and upload.
  • Testing the ideas against real ad results, so that estimates stop being guesses. We have no results yet.
    • You can cancel a paid plan in Settings with one click, no email or call needed. The plan runs to the end of the month you paid for, nothing more is billed, and you get a confirmation number. Until then you can keep it with one more click.
    • Download my data in Settings makes a file of your projects, product profiles, ideas, canvases, sessions and file links on the spot.
    • The site has a new header with Features, Models, Resources, About and API, and a page for each.
    • A models page lists every model on the canvas with its maker, speed and price, read from the same list the canvas uses.
    • Resources has guides on briefing your product, picking a video model, testing your ads, and safe zones and AI labels.
    • Ad packs for testing: distinct concepts, three hooks each, copy, a static ad, file names and a test plan, on the ideas page and in the chat.
    • The home page shows clips that model makers publish themselves, each credited and linked, and says plainly that none was made with Genien.
    • The company behind Genien is now named on every page: Vital Antioxidants LLC, a California company.
    • Models on the site now show their makers' logos, and every guide opens with a picture of what it teaches.
    • The plan is now written three ways and scored on hook, story, picture against words, product and what the models can really make, before anything is generated.
    • Each shot makes two frames at once, and the check keeps the better one. A redo happens only when neither passes.
    • The product's photo now goes into every clip, and every frame and check names its parts and colors.
    • Clips are made with MiniMax H3 Max by default. Shots where only the camera moves keep the whole scene frozen, so nothing in them can drift.
    • The script and any later rewrite now steer clear of what a general AI suggested for the product.
    • A video now costs about $3 to $5 of generation, a little more than before, because of the extra frames and checks.
    • Each flagged step in a video's report now has a Fix button: the check's new prompt goes on, and the step and the edit run again after you see the price.
    • Genien now picks up to three clean views of your product from its photos and leaves out infographics and room scenes.
    • Plans no longer ask AI to show how a product works from a single frame. They also say which moment to film on your phone as proof.
    • Storyboard first: a run can make only the frames, so you check and fix them before paying for the clips. Then one button makes the video. You can also choose it after a yes in the chat.
    • A product that folds, opens or cleans something is shown as a before and after: a first frame, a last frame made from it, and a middle one when the change has several steps. You check them in the storyboard before the clip is made.
    • If your product folds or opens, add a photo of each state. Genien finds it and gives it to every frame of the change.
    • Each kind of frame now goes to the model that makes it best: GPT Image 2.5 for products, hands, people and fine detail, and Nano Banana 2.1 for the empty room and for phone-style videos. Every frame says why it got its model.
    • Frame prompts follow the rules the model makers publish: full sentences in a set order, exactly one product, and no filler words such as 8k.
    • A frame that's right except for one detail can be fixed in the chat: Genien edits that picture and keeps the rest.
    • A frame guide shows what photos to give, how to check and fix a frame, how to show a change, and which model makes which frame.
    • A public run log lists every test run that reached a video, with its time, price, checks and flags.
    • A video with an AI voice now carries the full synthetic-performer label, and the label is a little larger.
    • If a check can't run on a file, the video's report now says so.
    • Every video now ends on an end card with the ad's closing line, set in a font so it's always spelled right.
    • Frames now look at one empty room instead of the first frame. Later shots used to copy its camera and pose.
    • Each shot now sits under its own words. The plan says which words go with which shot, and the cuts follow them.
    • You can close your account in Settings. Projects, files, videos and your sign-in go with it.
    • You can rename and delete a project in Settings.
    • The chat on the canvas can tag any node with @, remembers the conversation, and can be stopped.
    • Both chats start on Auto, which picks a model for each message. You can still pick one yourself.
    • Running several steps now always shows their price first.
    • Three voices now read each script. Pace, pitch, noise and clicks are measured, and you can swap the voice with a click.
    • Models now watch and listen to images, clips and voice before the edit. Every video gets a report of what was fixed and what's still worth a look.
    • First finished video from start to end on a test product.
    • Product card from a link, with ideas compared against what a general AI answers.
    • Each idea you make keeps its estimate, with a timestamp and a fingerprint, and you mark how it did.
    • The whole product is now in American English.
    • The canvas gained 36 tools and about 100 models, and you can edit a canvas with words.
    • A run keeps going when you close the tab.