Getting Started with the AI Image Generation Tool
The AI Image Generation tool is located on the left of the homepage - selecting this will take you through to the area where you can create images and assets.
There are various features and tools to take advantage of:
Located on the left of the image prompt text input, this fun and useful feature allows you to effectively 'roll the dice' and have the platform create a random and unique prompt for you. In addition, you can use the same tool to improve your own prompts for better model understanding, providing it's under 200 characters.
For more help on understanding prompt structure, check out the Beginners Guide to Prompting.
Prompt Magic V2 (& V3, Premium only)
This feature significantly improves prompt adherence and enhances image composition when combined with any model on our platform. Prompt Magic V2 can be found within the sidebar, while Prompt Magic V3 can be found in the Alchemy settings when Alchemy is enabled (Not available for SDXL models).
High Contrast (Prompt Magic V2 only) is active when Prompt Magic is turned on and produces increased contrast and shadow in the output images.
Prompt Magic Strength
This slider controls the intensity of its influence on the output, allowing you to experiment and achieve different results. We suggest trying values between 0.3 and 0.4 for optimal results.
Prompt Magic V2 is available for free users whereas V3 is a premium version that is usable in conjunction with Alchemy.
Leonardo PhotoReal represents an iconic feature within the Leonardo.Ai suite, offering a sophisticated image creation pipeline. This feature specializes in generating highly realistic visuals, suitable for a range of applications including cinematic scenes and authentic-looking portraits.
Depth of Field (PhotoReal only)
This controls the depth of field of the image (effectively how much is in focus). At lower values more of the image will appear soft or blurry. At higher values, the image will have a wider focal range, and more of it will be in focus. Experimentation with this option can yield unique and engaging results.
Raw Mode (PhotoReal only)
Enabling this mode will give more control and flexibility with longer and more complex prompts. Leaving it off is preferential for shorter prompts.
Choose from 3 different presets: Cinematic, Creative, and Vibrant. Each with their own unique influence on the generated output.
Leonardo Alchemy V2 is the latest version of our celebrated signature 2D image pipeline, which enables incredibly high-fidelity image generation and coherence for SDXL models.
Alchemy V2 not only improves prompt adherence and image coherence, it allows for prompt weighting (not available for negative prompts), making it the preferred choice of tool on the platform.
Alchemy V1 is still available for Stable Diffusion 1.5 and 2.1 models, and comes with several settings to experiment with. Please see below for an explanation of the various settings and options.
High-Resolution (Alchemy only)
This boosts the output resolution of Alchemy. High-Resolution outputs differ from normal outputs due to a unique diffusion process; it's not simply an upscaler - it can increase the quality of images, but it might lower prompt adherence.
Expanded Domain (Alchemy only)
When activated, Alchemy users may experience stronger prompt adherence, but it can also lead to occasional visual artifacts (unwanted aspects that are out of place, like a floating limb, or chunk of building etc). When deactivated, images are more likely to be aesthetically pleasing, but may not be as prompt adherent.
Contrast Boost (Alchemy only)
This feature adjusts the dynamic range of images when using Alchemy, enhancing the overall contrast - (Dark blacks and bright whites etc) This is a great option to play with, particularly when going for a certain theme, such as ‘gothic.’
Resonance (Alchemy only)
Resonance determines the level of detail and prompt adherence in a generated image relative to prompts. Having this setting high is tempting, but it can lead to unwanted positioning or artifacts. At the same time, having it too low means that the model could be given too much freedom, leading to random results.
Elements offers users the ability to finely tune the style and details of images using a 'mix and match' implementation of dynamic style modifiers. Located just below the image prompt bar, users can activate Elements and choose up to 4 that are compatible with the selected model, producing unique and customized artwork with an almost infinite amount of range and versatility.
(It is important to note that not all Elements work with all models – those that can't be used will be either greyed-out or not shown).
This vast upgrade to the popular ControlNet feature is designed to refine image style and composition with greater precision using reference images as a base. It introduces a large suite of new baked-in tools, including multiple image uploads, independent weighting, and simultaneous use of up to four ControlNet options for premium users.
Image Guidance has multiple pipeline features that create unique and varied results on produced images, such as Depth, Edge, Line Art, Normal Map, Pattern to Image, Pose to Image and QR code to Image (among others).
Using Image Guidance is easy and straightforward – it can be activated by selecting the option next to Generation History below the prompt bar and uploading a primary image.
(Note that some Image Guidance options and the use of multiple reference images requires a premium subscription plan).
This is where users can choose various sizing options for the images created, using either pre-selected dimensions, an aspect ratio selector, or a slider tool for width and height if they want a more unique scale.
It is important to note that certain models are trained at pre-set dimensions and can produce unwanted results if used for dimensions outside of this. A yellow warning sign (⚠️) will appear if the size is incorrect, whilst a star (⭐️) will be placed on the optimal ratio for the trained model (Not applicable for SDXL).
This is a slider scale from 1-20. It dictates how heavily the words used in prompting are weighted in creating the final result. It’s important to note that guidance that is too high or too low can result in unwanted aspects of a generated image. (We advise between 7-9 for the best results, but users should experiment to find what works best for them).
This option is great for those wanting to create textures (such as walls or floor tiles) and as such is ideal for environmental aspects in video game assets. Other possible applications include patterns for prints and as an pattern for wallpapers on your devices.
Users with paid subscriptions have the ability to make their images private or public. Any images kept private remain the sole property of the creator, while those that have been shared publicly can be used by both other users and Leonardo.AI in any capacity.
Please note that users cannot switch an image from public to private or vice versa once it has been created, with the exception of recent generations in the AI Image Generation page via the ... context menu, and that this feature is reserved for subscribed users only. Read more here.