Pick your AI model, and make thumbnails that cut you out on your device

What's new in v1.9.131?

• Choose the AI model in each AI tool: GPT 6 Sol (default), GPT 5.6 Terra, GPT 6 Luna or GPT 6 Astra for text; Kling v3 (default), Veo 3.1 or Seedance 2.5 for AI video; Nano Banana Pro (default), FLUX.2 or GPT Image 2 for images; ElevenLabs v3, Multilingual or Flash v2.5 for voices.
• GPT 6 Astra and Veo 3.1 are part of Full Access. Signed in on the Hobbyist plan? You get one free try of each.
• If a newer model is busy or fails, the tool falls back to the proven older model, so your job still finishes.
• New AI Thumbnail Composer: put a title and subtitle on any AI background, with colour, outline, shadow and size controls you can drag into place.
• Use current frame: start a thumbnail from the frame under your playhead. The frame stays on your device.
• Face Cutout: cut yourself out of the current frame right on your device (nothing uploads) and place yourself on the thumbnail with a white sticker edge, a soft glow, or no edge.
• Thumbnail Compare: save two versions as A and B, see them side by side, then download the winner or add it to your timeline.
• Redesigned home, pricing, FAQ and How it works pages, with a live proof panel that counts the video bytes uploaded while you edit. It stays at 0.

A quick way to build a thumbnail from the frame you already have

A good place to start with this update is the simplest workflow: pause your video on the exact frame where your expression, gesture, or subject is strongest, then use that frame as the base for the thumbnail. That matters because a thumbnail usually works best when it matches the real video instead of feeling like a separate poster. If you already have a usable moment in the edit, you do not need to rebuild it from scratch with a generated image. In AetherCut, you can begin from the frame under the playhead, then layer text and a cutout version of yourself on top. Because that frame stays on your device, this is also the fastest route when you want a custom thumbnail without sending the source shot anywhere else.

One practical approach is to treat the thumbnail like a three-layer composition. First, choose the base frame. Look for a frame with a clear subject, some empty space for text, and enough contrast between you and the background. Second, decide whether the frame needs a face cutout. A cutout is useful when the original frame is busy or when your face blends into the scene. Pulling yourself forward with the white sticker edge or a soft glow can separate you from the background without changing the underlying image. Third, add the title and subtitle only after placement feels right. A common mistake is writing the text first, then trying to squeeze the image around it. It usually works better the other way around: lock the visual focal point, then move the text into the remaining space and adjust colour, outline, shadow, and size until it reads cleanly at a glance.

Another useful habit is to compare two very small changes rather than two completely different concepts. For example, keep the same frame and same text, but make version A with the sticker edge and version B with the soft glow. Or keep the same composition, but test a shorter title against a more specific one. That is exactly the kind of choice that is hard to judge from memory and much easier to judge side by side. When both options are visible at once, you can usually spot which one feels clearer, less crowded, or more balanced in a second or two.

If you want to involve the new model picker, use it where it actually helps instead of changing models just because you can. For thumbnail work, the biggest gain usually comes from image generation when you need a background that your timeline does not already provide. If your own frame already tells the story, start there and keep it simple. If you do generate a background, it is smart to have the text and cutout plan in mind before choosing a model, so you do not end up with a detailed background that fights with the headline. And if a newer model is under load, the fallback behaviour means you can keep moving instead of babysitting the job.