Parameters

Parameters sit between the user’s intent and the model’s generation process, helping to guide how the AI interprets the input, weights different aspects, and commits to an answer. Essentially, they act as the knobs on the machine that let users control how tightly or loosely the AI behaves, how exploratory or constrained it should be, and how much initiative it should take.

The earliest form of parameters was raw flags typed into prompts. Midjourney popularized this with inline commands like --v 4 to switch model versions or --no dogs to exclude unwanted tokens.

Since then, this pattern has become more user-friendly. Parameters migrated into interfaces: sliders for temperature, dropdowns for model versions, quick chips for tone or length.

Parameter forms

Parameters are used across all content formats and use cases.

• Inline flags: These include typed tokens inside of the prompt. They offer high precision, but also a high learning curve. Examples include Midjourney’s --no to exclude tokens, --v for model version, or --ar for aspect ratio. These can be added manually, or reflect in the prompt box after being added from within the UI.
• Toggles and switches: When parameters have only two values, a toggle makes adjusting it easy. Copy.ai offers a toggle to make the writing more casual or more formal, and Jasper provides the option to bias for speed vs. quality. This parameter can also impact model behavior, such as the option to enable linting by coding copilots.
• Sliders: To dial up or down the weights associated with different parameters, sliders have become a common UI option. These can be open-ended or have set tick marks to represent common values. For example, OpenAI has recently introduced sliders in their canvas tools to control reading level with set levels to choose from, while Midjourney's OmniReference weight is open ended.
• Matrix controls: To add an additional dimension, two related sliders can be combined in a 2x2 matrix, granting more granular control over the balance between the two. The tradeoff to this heightened precision is a more complicated interface. Figma Slides uses this UI when using AI to adjust voice and tone.

Location and visibility

Parameters may become critical tools for more advanced users, but people who are just learning your interface and product may find them distracting or confusing. Balance visibility with ease of access.

Options that control cost or speed of generation, or the general format of the final generation such as length or aspect ratio may need to be more visible or even available by default from the input field.

Group more complicated parameters in panels and consider how progressive disclosure or relative association might be utilized to expose advanced options at the moment they apply.