Text to image - Walkthrough
Welcome to the very basics of AI image generation. This is where many services stop, but where our work just begins. Text-to-image generation means you enter a description of what you want to see into a text field (this is called your “prompt”). Then hit create and wait for your machine to visualize exactly what you had in mind.
The Process
This tutorial could, of course, end here as there is nothing more to say about that.
However, there is a lot more to explore because no machine can perfectly understand your needs just from your words. We provide various options to influence the generation before hitting the “create” button.
You may have seen the image of the tree before, but it is necessary to understand what happens here. You have influence over every step up to the final output (in other words: your picture).
Choosing a Model
First, choose a model that matches the style you want to achieve. If you want to learn more about models, read our “train your own model” and “bring your own model” tutorials.
To choose, click on the left side on the name of the model (f.e. Flux Dev, SDXL, ...). You will find a variety of different models listed. Select one by clicking it. Then continue by using the promptfield right next to it.
Using the Prompt Support
Underneath the promptfield you will find the "Prompt Support". There you have the option to add keywords to your prompt. Depending if the model supports "Negative Prompts" you will also find the field for it here. And we have added an "Image to Prompt" option.
Negative Prompts
The negative prompt is where you insert everything that you DON’T want to see, for example, “deformed hands,” “3 eyes,” or “blurry image.” Remember: There are use cases where these things don’t have to be excluded. A blurry image can be a core style choice for a perfect image. That is why we don’t switch these things on automatically. Negative prompts are parts of f.e. SDXL basemodels. For Flux this has no impact and therefore is not displayed to use.
Settings Overview
Model Control
We hided some of the settings that weren't used very often. But you can still access them. F.e. the number of generated images is now up to 6. But keep in mind that the render time will be longer depending on the amount of images generated. Here you can also select the aspect ratio of you images
Lab Mode
we decided to go back to the original names of the open source community so that experts can use these settings the way they are familiar with them. Still we kept it low profile as these settings are usually not noticed by most users. Still there is a little explanation to each of these settings:
Sampler
You can experiment with the sampler as it has a significant impact on the generation, but if you are unsure, just keep it on default.
Steps
The steps refer to the number of steps required to generate an image. Think of it like creating a painting: the initial step provides a rough sketch, and with each additional step, the image becomes more detailed and precise. The quality of the image is therefore linked to the time invested in these steps. However, there is a point where further steps add no noticeable improvement. We've found that a range of 20 to 40 steps is typically sufficient, as increasing the steps beyond this range doesn't significantly enhance the image.
CFG-Scale (Classifier Free Guidance)
The CFG-Scale lets you control the level of creativity the AI uses, allowing you to decide whether it should follow your prompt closely or take more creative liberties.
Seed
The Seed can be thought of as the origin of your image, like the seed of a tree planted in the ground. If all conditions remain the same—same light, same amount of water, wind, temperature, etc.—the same tree will grow again. By hitting the lock image on the left, you can fix it. Otherwise, it will always randomize it.
Now you are all set and ready to hit the “create” button next to your prompt. A couple of seconds later, your image should appear.
We have tutorials on even more features in our blog-section. Have fun exploring our just go directly to Inpainting or background-removal