2.2606.302

New Features

• Portrait / landscape orientation for AI image & video generation: a new orientation picker (Portrait 9:16 / Landscape 16:9 / Square 1:1, portrait by default), suited to mobile short-video and other vertical scenarios; the chosen aspect ratio is passed through to the upstream model with the generation request.

Fixes

• Fixed artifact download failures for built-in AI image / video / voiceover — generated results now download and save locally as expected.
• Fixed an intermittent 503 / no available upstream error in built-in chat under certain built-in provider configurations.
• Video generation now correctly forwards the random seed, negative prompt, motion strength, and camera-motion parameters.
• Fixed a rendering (hydration) warning on the app logo in some scenarios.

Improvements

• The image / video generation toolbar is now responsive: on a narrow panel, secondary controls (orientation / count / multi-view / advanced params) collapse into a "⋯" menu, and the model selector is no longer squeezed or truncated.
• Controls available for built-in generation models (reference image / seed / negative prompt / keyframes / camera motion, etc.) now show or hide based on each model's declared capabilities.
• More accurate UI language detection: full language tags reported by the OS / browser (e.g. zh-Hans-CN, zh-TW, en-GB) are normalized to a supported UI language instead of incorrectly falling back to English.
• The new and improved UI copy above is localized in all 5 languages.

changelog