Ask Aius for what you want in plain words. Aius chooses a model that can do the job and tries another if a provider fails, or you choose the model yourself (see Choosing the model). Every use case below is shown in a short recording on the Videos page.

Choosing the model

Every media model in /models and on Prices has a one-line description of what it is good at, and a price tier within its category. See Finding a model for search words and what to ask Aius. There are three ways, and the first one that applies wins:
  1. Name it in your request, for example “make it with Nano Banana”, “say it with ElevenLabs v4” or “transcribe with ElevenLabs Scribe”. Any model listed for that kind of media on the Prices page works.
  2. Choose it in /models, in the terminal or in AIUS Light. Recommended shows AIUS’s picks for each kind of media; pick a category such as Image editing or Speech, or All, to see every model of that kind with its price. Typing a search, such as “elevenlabs” or “flux”, finds any media model. Your choice is kept on that computer until you pick Auto again. It works with whichever chat model runs the conversation, Pareto included, and the terminal confirms which model that kind of media now uses.
  3. Auto (recommended): Aius picks a model that does the job well at a fair price, and tries the next one if a provider fails before it starts.
Every provider’s media models are included, and each one lists every provider that serves it, in the order they’re tried: OpenRouter’s whole list, plus OpenAI’s own. OpenAI’s image, transcription and embedding models run through OpenRouter first, then directly on OpenAI. TTS-1 and TTS-1 HD, which only OpenAI serves, are among the Speech models. Moderation checks text with OpenAI’s Omni Moderation, at no charge: ask Aius to “check these comments for anything harmful”. Live voice conversation models (OpenAI’s realtime and gpt-live) aren’t available yet. Pointers named latest are never used: every model is pinned to a dated release, so a model and its price don’t change under you. A model you named that can’t run says so; Aius doesn’t quietly use another one. Ask “which voice models do you have?” to get the current list with prices and voices. A chat model that can’t read images, such as Pareto, still answers questions about images and videos: Aius has a vision model look at the image, or at frames sampled from the video, and passes on what it saw and the exact text.

Where the files go

Aius writes each file into a media/ folder in the folder you are working in, with the date in the name, and never overwrites an existing file. media/.aius-media.jsonl records each file’s model, provider, cost and time. When the folder is linked to an AIUS Light project, light also stores the file with the project, together with the files it worked from (the recording it transcribed, the photo it edited, under media/sources/). Everyone in the project sees them in AIUS Light. Nothing else in your folder is uploaded.

What it costs

Each result says which chat model asked for it, which model made it, the provider and what it cost, for example “Aius on Pareto running Nano Banana 2 via OpenRouter · $0.08”. The cost counts in the answer’s total and in your organisation’s budget, like chat, so you see what the whole request cost: the chat model’s work and the image, speech or video together. Every media model and its current price is on the public Prices page, where you can search and filter by category, price tier and lab without signing in. Each model has a line on what it is good at. Prices are in US dollars with AIUS’s 25% margin included (the provider’s price divided by 0.75), per image, second of video, minute of audio or 1M characters, and follow the providers’ own price lists. Price tiers (Cheap, Affordable, Expensive, Premium) compare models of the same kind. A video is charged only once it is ready: a video that fails costs nothing. To choose the model for each kind of media, open /models: image generation, image editing, speech, transcription, video, embeddings and reranking each have their own tab. See Models. Aius asks first before any video, before more than ten items at once, and before anything it estimates at over $0.50.

Private inference

If your organisation uses private inference only, media requests never leave it. A kind of media your private deployment can’t make says so instead of using another provider.

In AIUS Light

Open the conversation in AIUS Light: each image appears in the chat (click it for full size), speech and recordings have a player, videos play inline, and transcripts show as text, each with the model that made it and what it cost. The same files are in the project’s Files view under media/, where you can preview, download, rename or delete them. In the terminal, each result shows the file’s path in media/, its size and “in AIUS Light” once it is stored there.