Modalities

The Studio is one conversation, not six separate screens. In Manual mode, opening the model picker starts with What do you want? — a row of chips for the output family (Chat, Image, Video, Audio, Music, Transcribe, …) — and picking one swaps the composer’s input controls and output renderer to match. Your project, chat history and credit balance stay the same throughout; only the controls around the message box change. This page had no picture of that chip row before, so here it is:

​

Chat

Text-in, text-out (or text with images/audio attachments). Supports streaming, tool calls, JSON mode and vision on capable models. Model picker shows every chat-capable model. For vision, attach an image. For PDFs, models that support them read directly; others get auto-converted text + page images.

​

Generate Image

Text prompt → image. Controls: aspect ratio, number of images, style (model-dependent). Models like Nano Banana (Gemini Image) additionally support image edits — attach a source image and describe what to change. Output is stored in GCS, shown inline, and re-attachable to chat messages with one click.

​

Text-to-Speech

Text prompt → MP3/WAV. Controls: voice, speed, format. The audio is streamed back as a binary response (or HTML5 player in the UI).

​

Speech-to-Text

Drop an audio file or record directly in the browser. Output: transcript (with optional segments and word-level timestamps on verbose_json format). Accepts up to 25 MB audio. Transcript appears inline and is searchable in the chat history.

​

Generate Video

Text prompt (and optional reference image) → MP4 up to 16 s. Controls: duration, resolution, aspect ratio, people allowed (for Google Veo). Video generation is async — the UI polls the job and shows progress 0 → 100 %. Completed videos are mirrored to our storage so they survive past the provider’s ephemeral URLs. See Create video generation for API details. (Mirroring is a Studio behaviour — a video generated through the API is not mirrored and its URL expires with the provider’s.)

​

Generate Music

Text prompt → full song (vocals, lyrics, instruments). On Suno models you also get a style builder and a vocal-gender choice; length isn’t a separate control on any provider — it follows the prompt and style rather than a slider. Takes 30–90 seconds typically; streams progress updates.

​

Picking a modality

Switch mid-chat — changing the family doesn’t reset your chat thread. You can generate an image, then switch to Chat and ask the model to describe it, then switch back to Video and animate it. All with the same credit balance.