When using Text to Image to generate an image using FormAI, the Model Options are important settings that will have a big effect on the final results you see and how many credits you use in generating your image.
Model Options
The first and likely most important selection is which Model you would like to use. The various models use different data sets and will interpret your prompt uniquely.
- To locate Model Options, navigate to FormAI, select Images + and then select Text to Image.
- You will find the Model dropdown on the top right-hand side to select for image generation.
Lumo
Lumo is our latest model delivering the highest-quality images with an unprecedented understanding of user prompts. It is also very good at rendering text.
- Example prompt: A man making a wooden table. Attributes used are DSLR camera and cinematic lighting.
Astra
Astra is an extremely flexible model, able to understand complex, multi-subject prompts, such as a man wearing a blue shit or a woman wearing a red dress.
- Example prompt: A man making a wooden table in his garage at night. The attribute used is a 35mm camera lens.
Felli
Felli supports Mixers, which enables you to integrate consistent faces, objects, art styles, or even pets that the other models may not be able to reproduce.
- Example prompt: A man making a wooden table, mixer used an attribute in Van Gogh's art style.
Please note: Mixers need to be “trained” by uploading a series of images of the subject you are trying to reproduce. The minimum number of images to train a mixer is 20. If you can add more images, that will help in the training process.
Quadra
Quadra excels at high-resolution lifelike images. Also supports an engine that forces images to look a certain way. Examples would be Anime, made out of clay, or pixel art.
- Example prompt: A man making a wooden table. The attributes used were scene and sunny.
Please note: The engine dropdown will appear underneath the model dropdown when Quadra is selected. The engine used in this example was Cinematic.
Norxa
Noxra is our least expensive model. It understands a large variety of inputs and produces extremely high-quality photorealistic images. When selecting Noxra, the quick prompt Photorealistic is applied as well, selecting the appropriate camera, art style, and technical filters for the best image possible in this model. If you choose to not use these preselected fields, you can clear filters and select your own.
- Example prompt: A man making a wooden table. Attributes are 35mm camera, photorealistic, photograph, and photorealism.
Fresko
Fresko has been trained to render text. It also excels at logos & line art.
- Example prompt: Create a logo for a local baseball team. The attribute used is Concept Art.
Please note: It may take several attempts to accurately render text using Fresko.
Model Option Settings
Under the Model Options section are several settings that mostly play into how many credits each image you generate costs. Not all options apply to every model.
Pro tip: If you are finding the model is generating elements you did not expect, try adding the things you want to exclude in the Negative Prompt field.
Quantity
Quantity determines how many results the model generates based on your prompt. While more images will give you greater variety, each image will cost credits based on the settings you choose.
Aspect Ratio
Aspect ratio is the proportional relationship between the width and height of the images you will generate. In general, Square (1:1) is the best option, as the AI tends to want to fill in the space inherent in other aspect ratios, causing unexpected results.
Size
Size relates to the physical dimensions of your image, combined with the aspect ratio. For example, selecting an aspect ratio of Photo (5:4) and size of Medium (512px) will result in a 640px wide by 512px tall image. Larger images require more processing and therefore cost more credits.
Please note: When displaying images in your browser, images may be scaled to fit. Once downloaded they will be saved to your computer in their true size.
Generation Steps
Generation steps is a technical term for the number of iterations the model takes to generate and refine an image. More steps may help with detail but after a certain point, may not necessarily improve the image. In general, around 30 steps (the default) produce an ideal image.
Seed
The seed is a random number that the model uses as a starting point for generating an image. Each image has a unique seed, and if you want to reproduce an image or create variations, you’ll want to take note of the seed. To locate the seed, select the picture you want to use as your starting point by selecting the three dots and then copy the seed to the clipboard.
Prompt Strength
Prompt Strength (sometimes referred to as CFG Scale) adjusts how closely the model will render images to your description. Lowering this number from the default of 7 will increase variety in your results, but may not look like what you were expecting.
