Skip to content

Commands Overview

Parameters allows you to take the level of control of your image generation process to the next level. Parameters can be added to your prompt text in a specific format (typically --parameter_name format) and each of them does something to the image generation process. Make sure to try each of them to achieve the next level control!

This is a list of parameters available for /serve and /serve-free commands:

Parameters:

Prompt

  • Usage: Simply start your command with the desired image description.
  • Description: Describes the kind of image you wish to generate. You don't need to prepend --prompt; the system assumes your first input as the prompt.

Negative Prompt (--neg) Learn More

  • Description: Indicates what you'd like to exclude from the image.
  • Example: --neg light bulb

Seed (--seed): Learn More

  • Description: Set a seed for your image. If unspecified, a random seed is selected. Useful for recalling a specific image or comparing parameter impacts.
  • Default Value: random number
  • Example: --seed 1991

(CFG) Classifier-Free Guidance (--cfg) Learn More

  • Default Value: 6
  • Description: Dictates the model's adherence to your prompt. High values can impact image quality.
  • Example: --cfg 8

Aspect Ratio (--ar) Learn More

  • Usage: Specify as x:y (e.g., 1:1, 9:16).
  • Description: Changes the generation's aspect ratio. Available options vary depending on the model (SDXL or SD1.5).

Denoise Strength (--power) Learn More

  • Default Value: 0.3
  • Description: Adjusts denoise strength during upscaling. High values can introduce artifacts in images.

Model (--model)

  • Default Value: 'cosmopolitan'
  • Options: 'sdxl', 'vodka', 'screwdriver', 'mohawk', 'bloodymary'
  • Description: Chooses the model for image generation. Note that mohawk is SDXL custom model and SDXL workflows are applied

LORAs (Low-Rank Adaptation) (--lora) Learn More

  • Description: Appends an optional model for enhanced generation. For 'bloodymary', 'vodka', or 'screwdriver', there are dozens of LORAs available. SDXL currently supports 'nurlens'. See all othe LoRAs available via /command-infos and here.

Control Image (--control)

  • Usage: Provide a Discord-hosted image URL.
  • Description: Constrains image generation based on the outlines of the provided image. Can be used with --image.

Image to image (--image) Learn More

  • Usage: Provide a Discord-hosted image URL.
  • Description: Uses the specified image as a starting point for generation.

Inpainting (--inpaint) Learn More

  • Usage: Provide a Discord-hosted image URL that contains masked area in alpha channel.
  • Description: Edit the image using a mask. Generation will happen only on the masked area
  • Note: Generate masks using 3d party tools or with /makemask command in Distillery. See details in Learn more.

LLM: Large Language Model Assistants (--llm) Learn More

  • Usage: Transform your simple prompts into more interesting versions
  • Description: Fine-tuned LLMs are applied to user prompts to alter them in various ways

IP Adapter (--adapt) Learn More

  • Description: The --adapt parameter, allows user to prompt Stable Diffusion with the image input. It is a powerful tool that can be used in a various workflows for style, charachter, or other levels of control.
  • Usage: Provide a Discord-hosted image URL. up to four IP Adapters can be used per generation: --adapt, --adapt2, --adapt3, --adapt4
  • Note: Read documentation for the details, --adapt feature has a bit of learning curve.

Sharpen Tool (--sharpen) Learn More

  • Usage: Makes images 'sharper', users can adjust the amount applied to images
  • Description: Post-processing tool that improves image aesthetics.

Sampler Selection (--sampler) Learn More

  • Usage: Choose between legacy and LCM samplers "lcm", "legacy"
  • Description: Allow selection of the sampler.

Raw (--raw) Learn More

  • Usage: add --raw parameter to the prompt to disable additional processing
  • Description: For generations that require higher level of control.

Upscale Method (--upscalemethod) Learn More

  • Usage: Allows selecting which upscale method will be used: "legacy", "direct", "none"
  • Description: Impacts image quality and generation time