When using the image generation function you can use these features to generate the image you’re looking for.
/w=3840,quality=90,fit=scale-down)
Selected model
Here you can select from the available AI models that will Dream up the image you’re looking for. To help you choose the model that works for your project you can go through some examples on the choice screen.
Prompts
In this section you should describe the end result you wish to see generated. Be as specific as you can but don’t go against your input picture. For example you shouldn’t just prompt a command like “Fill it with furniture” when your input image is an empty room or office. Instead you could describe the pictures as you’d like to see it using simple sentences or words like: Open office layout, modern interior, cabin at the edge of a forest near a lake etc.
To help with this you can use the auto generate to let the AI tell you what it thinks about your input picture.
If there is something on your picture that you do not wish to see you can use negative prompts to get rid of them the same way as you’re using prompts. Simply describe the thing that shouldn’t be in the picture.
Input image
/w=750,quality=90,fit=scale-down)
There are two ways you can add an input picture to your generations.
Firstly if you have installed the plugins for your DCCs, you can grab a viewport directly through Project Dream from them.
The supported hosts are:
3dsMax, Revit, Rhino, SketchUp, Blender, Cinema 4D (Full list, and details of supported plugins)
You can also upload an image directly by drag and drop or uploading it from your computer.
After adding the image you can use the image mask tool to specify which parts of the picture you want modified. Unmasked parts will not be modified.
The creativity slider lets you choose how much you want your image altered. Higher percentage lets the AI alter the image more. This could mean more detail and better results but this can also make the picture come out less like the original.
The image strength slider tells the AI how closely to follow the input image. Higher percentage means that the outcome’s composition will more closely follow the input image, but might introduce artifacts at maximum value.
Reference mood
Uploading a reference picture lets you set a mood you’re looking for. You can use this to give your picture a setting, a time of day, a certain season, just general colors or artistic style.
Reference type lets you choose how the style will affect your end result.
Light: Generation should consider the reference images style and color use more heavily while ignoring the actual things depicted by the image.
Medium: This option let’s you add some of the stronger details from your reference image on top of it’s style and use of color.
Heavy: Using this option will consider the things on your reference as much as it’s style. Using this option should make parts of the reference bleed into your end result.
The influence slider sets how much the outcomes mood will be influenced.
Settings
If you don’t have an input picture and you’re only using a prompt you can choose one of the preset aspect ratios or you can set your exact height/width later in the advanced section.
You can also choose how many times you’d like to generate an image with the same settings.
Advanced settings
If you wish to go a little bit deeper while generating your image here you can find some advanced settings.
While using prompt only you are free to choose a custom image width/height if the preset ratios don’t work for you. Image width and height can only be set while not using an input picture.
Samples lets you choose how many iterations the AI will go through before the picture is considered finished. Higher samples can result in better quality pictures but will take more time.
Seed is usually a random number that makes it possible to have different outcomes with generally the same settings. You can set a custom seed if you only want to change other settings, but keep in mind that the same seed and inputs will always gives the same end result.
You can think of the sampler as the image generation algorithm. Experiment changing this, will result in different flavours of inference.
The magic salt toggle turns our preset positive and negative prompts on or off. Power users can experiment with turning this feature off, and come up with the full prompt without any hand holding.