Parameters change how a response is generated by AI. They are closely connected to filters, but whereas filters constrain inputs and outputs themselves, like the source references or the format of the final product, parameters affect the weighting of different elements like tone, tokens, styles, references, etc.

Originally, these were hand written into token-based inputs like Midjourney. You might add --no [token] as a parameter to exclude certain tokens from the AI's synthesis. Or, you might use --v to change the model that the AI was using on the fly.

This pattern has rapidly evolved into something much more user friendly, relying on a combination of traditional UI patterns like defaults and selectors, plus natural language. The tradeoff of this user friendliness is decreased control, particularly in token-based generators. More savvy users may be able to guide the AI through natural language processing, but less experienced users may find it more difficult to get the AI where they want it to go.

Midjourney's Alpha web interface makes parameter controls more discoverable and easier to adjust, but at the cost of constraining which parameters are easy to find.

Parameters are used across all content formats and use cases. Some products like Udio or Hypotenuse's image builder template follow the Midjourney model and show many parameters at once coupled with the initial prompt. Others like Jasper make the initial choice very simple, with a basic selector to let the user bias for speed or for quality in the AI's result. Text-based generators will often include parameters on tone, simplicity, and so on as nudges when a user is inpainting using inline prompting.

A collection of sample parameters in action (Midjourney web,,, Hypotenuse, and Udio)

Approach these as you would a common filter system. Depending on the use case, users may benefit from more options up front, or may need a more simple interface to get off the ground. The necessity of parameters are a function of many considerations. For example, the benefit of the parameters shown by Udio, the audio generator, means it's much more likely to find a fun and delightful result on your first try-clearly a growth hook that the app's creators find useful.

AI can be very effective when the user is able to command it easily while leaving enough space for the generation to take the user in unexpected areas. Parameters can follow the user's intent, throttling up how constrained the AI is based on the user's needs, and not defaults in the system or training data.

An example of using inpainting and adjusting the parameters with each iteration to create an editorial styled image

Details and variations

  • Parameters cover a broad list of constraints and options that can be specified to the AI
  • They cover all media types, from image resolution and variance in image generators, to tone and style in audio generators, to voice and structure for text
  • Combine parameters with filters that govern the reference inputs, and token layering to create different results
  • Include parameters in the metadata of images for easy recollection by users
  • Include parameters in token and prompt transparency, as parameters and tokens can blend together without the user knowing



Proactively guide the AI

Parameters make it easy to explicitly set guardrails for the AI, instead of relying on advanced prompting techniques. When combined with inpainting and other iterative prompt techniques, they help the user guide the AI through iterative generation and maintain the user's control on their intent.

Creative exploration

Changing parameters can have unexpected results, which can feed creativity and open up new paths for exploration. When constraints aren't necessary, combining unexpected parameters like tone and audience let you expand the horizons for the AI, much like creative brainstorming supports our human ability to think beyond the box.

Potential risks

Sacrificing ease for control

More advanced users, or users working with a more complicated intent, require the most control over the AI. Simplify the interface and number of parameters for early users or less technical use cases, but consider how to make parameters available and manageable for other situations as well.


The Midjourney input in Discord allows users to select multiple parameters including image ration, model, negative tokens, and more
Midjourney has built their standard parameters as options into the web interface
Jasper's speed vs. quality toggle is a parameter offered by default
Grammarly offers preset parameters for audience and other constraints
Hypotenuse lets users select a tone as a parameter for the outout
The hypotenuse input includes a selector so the user can choose to set a parameter to bias for depth or speed.
Adobe Firefly includes pre-set parameters along with token suggestions
ChatGPT now allows the user to set the model as a parameter on individual prompts
No items found.