Docs / Image generation: inferred descriptions

Image generation: inferred descriptions

We can generate an image from the normal text a character writes, without requiring a special image prompt. This is the default inferred description flow.

How it worksThe system looks at the character's response and decides what part of the scene to illustrate.
Why it variesThe model can choose different details from the same response, so visuals may drift between messages.
When it shinesIt is the fastest setup for creators who want a minimal prompt and let the model interpret the scene.
Best practiceUse a consistent image style so the generated artwork stays on-theme even when the literal text shifts.

What the guide emphasizes

  • No special Definition setup is required to use inferred image descriptions.
  • The model can derive the scene directly from the character reply, which makes the workflow simple.
  • Inconsistency is a known tradeoff: the system may choose different visual details or miss the intended character/story tone.
  • We recommend using an image style to keep output visually cohesive across turns.

Related image-generation settings

Direct Image Descriptions give the character explicit italics-based prompts that can be tuned separately from the normal response text. Hidden Descriptions hide those direct prompts from the user. Image Styles keep the visual language stable across messages so the same character feels coherent from scene to scene.

Practical takeaway

Inferred descriptions are a great default for simplicity, but they are less precise than a direct image prompt. If you want stronger consistency, pair the model with a clear Image Style and, when possible, define more explicit image examples with Direct Image Descriptions.