Image generation
How to give Kwern an image model and a key of its own, which models it offers per provider, and how the key is checked before anything is stored.
What it is
The model Kwern draws a post's picture with, and the key that pays for it. It is its own setup, apart from the Provider key your runs cost on: changing the model your runs use never changes who draws your pictures, and deleting the Provider key leaves it as it is.
Pictures are drawn by Google, OpenAI or xAI. A key from that provider is all it takes.
Set it up
- Pick the image model — Pick the model that draws your pictures. Each sits under the provider that serves it, and the key comes from that provider.
- Google: gemini-3.1-flash-image, gemini-3-pro-image
- OpenAI: gpt-image-2, gpt-6-luna, gpt-6-sol, gpt-6-astra
- xAI (Grok): grok-imagine-image-2.0
For the provider of the model picked:
- Google AI Studio: Open Google AI Studio — Kwern runs on a key of yours. Google AI Studio issues them in Google AI Studio, under Get API key. Open Google AI Studio
- OpenAI: Open OpenAI — Kwern runs on a key of yours. OpenAI issues them in the OpenAI platform, under API keys. Open OpenAI
- xAI: Open xAI — Kwern runs on a key of yours. xAI issues them in the xAI Console, under API Keys. Open xAI
- Choose the key — Use a key Kwern already holds for this provider, or create one and paste it here. A pasted key is stored encrypted and never shown again.
For the provider of the model picked:
- Google AI Studio: Check — Asking Google AI Studio whether this key can draw…
- OpenAI: Check — Asking OpenAI whether this key can draw…
- xAI: Check — Asking xAI whether this key can draw…
Only the models listed can be picked, so Kwern knows what each picture costs.
When Kwern already holds a key for the provider you picked, the key step offers it: your Provider key, or the key your image setup already uses. Reusing it copies the key into the image setup, so replacing your Provider key later does not change it.
The chat models that draw
OpenAI's chat models that draw pictures themselves, such as gpt-6-luna, are listed among OpenAI's choices. Picked there, the picture is drawn on OpenAI's image model with the same key, since an images endpoint takes an image model and turns a chat model down.
Turning it off
Image generation can be switched off in Settings without forgetting the model or the key, so switching it back on asks for nothing.
Products and surfaces
How a Kwern product is created from one URL, what a surface unlocks, and which documents every agent reads before it writes a word.
The MCP server
How to add Kwern's MCP server to a client, what it exposes, how your agent calls your connectors through it, and the calls that publish the Writer agent's articles from your own tooling.