Image Generation

Heiryn can generate images using AI models from supported providers. The effective setting can be an explicit model, Auto, or Disabled. Auto follows the personal, community, and deployment defaults that are available to the current request.

Available Providers and Models #

Provider and model availability comes from the deployment's provider templates. OpenRouter is public by default, but a deployment can change its visibility or model catalog and add more providers supported by its configured image executor.

Use /personal_image_model or /community_image_model to see the choices available to you. Autocomplete and dashboard selectors prefix each friendly name with its size category and store the size-prefixed alias, such as [S] Qwen Image 3 for s:qwen; the provider template maps that alias to the model ID currently sent to its API. The saved alias is stable, but its underlying model may be updated under the model alias policy.

The current public image catalog is generated from the deployed providers:

Only categories containing configured models appear in the table. Image size categories are relative catalog tiers, not output dimensions. The displayed catalog is filtered to the maximum model size available for the current scope.

Configuration #

Image models can be set at two levels:

• Personal: use /personal_image_model to set your own image model.
• Community: use /community_image_model to set the server-wide image model used by members without a personal selection.

An explicit model requires usable credentials for its provider. A configured provider key is one possible credential source (see Bring Your Own Key).

The optional provider parameter selects an available image provider and defaults to OpenRouter when omitted. The model parameter selects one of that provider's image models. Providers marked private are available only to operators; disabled providers do not appear. Autocomplete suggests models for the chosen provider or OpenRouter when no provider was entered.

Choose Auto (or reset with -) to remove the explicit override. Personal Auto continues to the community setting, then to the deployment default when one is available. Community Auto continues to the deployment default. Choose Disabled to stop image generation at that scope; a personal selection still takes precedence over a community Disabled setting.

How It Works #

When the AI determines that an image should be generated in response to your query, it uses the configured image model and provider to create the image. The generated image is included directly in the bot's reply.

If a model claims that an explicit image request succeeded without calling the image tool, Heiryn retries that turn with the tool required. The unverified claim is never shown as a successful generation.

Image generation and text responses can be combined in the same reply.

Image editing uses the same tool and selected model. Images attached to the current message are supplied automatically as references. When the current message has no image, replying directly to a message with an image uses that replied-to attachment instead. The number of references is capped to the selected model's advertised limit.

For follow-up refinements, Heiryn keeps the latest image it generated in the same conversation history. If no new image was supplied, the AI can explicitly reuse that image for requests such as adding, removing, or changing an element. It is not reused for unrelated new-image requests. Resetting or expiring the conversation history also removes this refinement reference.