Steps, number of images, and shortcuts for steps in stable diffusion

Steps refers to the number of render passes that the server takes to arrive at the final image. More steps can mean a more detailed image, while fewer steps can be considered a rough sketch, though in some cases less it can result in an “overcooked” result.

Usage:   /render /steps:25  and your prompt etc

25 steps is considered a good safe and fast number.  100 steps is the maximum that can be specified using /steps. However, you can hit 200 steps, explained below.

As seen on popular prompts on Civitai, you only usually need between 15-35 steps for a beautiful image.  There’s no need to do crazy step counts while you’re in your ideation phase, it’s just slowing you down. (see workflow below)

Not all steps are created equal

The sampler, the part of the software that removes “noise”, is ultimately determines how fruitful each step is.  A modern sampler like UPMS can do a lot more with a lot less steps, where as KLMS will need twice as many passes. You can lock your sampler of choice in your settings, or set it to random.

Less steps, more images

Unique to our software: Our community shares a large multi-server instance of Stable Diffusion. We want everyone to wait as little time as possible, so all renders take the same amount of time.  This helps us provide the most accurate time estimates, and beat them.

To balance this, we suggest using these shortcuts.  Each one takes roughly about the same time to render, but you can get more images back.

/render /steps:wayless — this command returns 9 images

/render /steps:less — returns 6 images

/render  — returns 5 images (default 50 steps)

/render /steps:more — returns 3 images

/render /steps:waymore — returns 2 images  (experimental, upper limit)

Suggested workflows

PLAN A:

Render without using /size or /images.  Instead, use the High Res Fix technique when you render something you like to get high quality images and variations at the same time

PLAN B:

We talk to many great AI creators — and they throw away 99% of their stuff. This is the way they work:

Begin with /render /steps:wayless to get a lot of ideas fast, then when you start to arrive at poses and images you like, use /showprompt

Showprompt is a command that gives you all of the information you need to regenerate an image.  Armed with the one you like, modify that prompt with more steps and you’re arriving at your destination faster instead of rendering slow responses every time that you may not want to keep anyway.  Showprompt is the fastest way to find out an image’s Seed.

Use:   Reply to an image and type

/showprompt

Number of Images

Our system is designed to return as many images as possible within the span of a minute to thousands of users at a time, so (at least at this time) there isn’t way to batch render 30 images at once yet.  We are contemplating expanding our queue system to allow multiple types of jobs, but this gives you an understanding of our challenges when setting up a command to handle the number of images.

As described in the commands above, the best way to juggle this is to use our /steps:wayless shortcut to get 9 images back.  Once you have 9 ideas, use /showprompt and then render that good result at a higher res.  You can also go about this manually using the /images:X  command

Example:  /render /images:9 a cool kitten

This returns nine 512×512 images.  If we ask for nine 1300×1300 images at 100 steps, it will take longer than 1 minute and another user will suffer long wait times, so this is not possible.  You’re free to experiment and find a sweet spot that works for you, but expect the queue system to fight back some.

You’re never supposed to see a message that asks you to be considerate, but if you push it, it will tell you so!

In the future we will offer a “CHAD MODE” where you can have multiple queue slots and render more images at once.  Please be patient while we scale this feature up.