About Civitai

Civitai has a lot of great features. Besides being an image generation site, Civitai is also a popular website for downloading user-trained Stable Diffusion models, which are needed by anyone running a Stable Diffusion interface on their own computer. As a learning resource, Civitai also displays a gallery of images showing what can be created with each model, along with the full prompt and settings used to create many of them. This makes it easy to see what's possible with a model and learn what kind of prompts it takes to get the best results.

If you visit Civitai.com and create an account, you are immediately given "Buzz" points you can use to enter your own prompts and generate images right on Civitai, using a wide range of models.

As soon as you log in, it's a good idea to find the drop-down menu in the upper right corner of the screen and choose "Account settings." There are important adjustments here regarding Content Controls and Content Moderation. Civitai lets people create uncensored images of all kinds of subject matter, so set these controls to determine whether you will see adult-oriented content before you start browsing the site. You have been warned.

After that, starting at Civitai's Models page, you can Filter models by different attributes:

• If you are planning to use the Create function (to generate your own images online) then you can search for "On-Site Generation." as the Model Status.
• Choose "Checkpoints" as the Model types to find the main models, rather than just helpers or accessories.
• You can also choose "LoRA" if you want a helper or modifier model, which will work as an add-on changing the look or style of your images, or providing specialized training on the look of a particular character or type of scene.
• Base models are the foundational models that were initially trained for each generation of models. The user-trained models listed on Civitai are all made as modified, enhanced copies of these foundation models, but still follow the same specifications such as base resolution. You probably want "SDXL 1.0" set here.
• Another choice for foundational model is something called "Pony," which started out as a SDXL, but has grown into its own strange ecosystem of model checkpoints and pony-compatible LoRAs.

Once you find a model that you like, click on its thumbnail. Along the top of each model page is a list of all of the versions that have been installed. The versions that have a paintbrush icon beside them are the ones you can use to create images online, directly in Civitai. Click on the version, or click the blue Create button at the top of the right column, and you're ready to use this model in your own images.

There's another creation option on the model page that might be even more useful to beginners. Scroll down to the Gallery section under any popular model, and you'll see lots of terrific looking images. Clicking on one will usually provide you with the full prompt and settings that were used to create it, and you can even press the Remix button beneath the prompt, copying all that data into the options for your own next generation. You can re-use the random seed to make a copy (or near-copy) of the image you just saw, or use a different seed to create a variant on that image.

Click the Create button at the top of the page, and choose "Generate images" (or visit civitai.com/generate) and you'll see the a column of image generation options on the left, with the rest of the screen full of all the current and former images that you have been generating.

The Generation column starts with the Model you picked (you can try your prompts with different models to see how differently they can come out), and a section for Additional Resources such as LoRAs that can modify or enhance your generations. Beneath that are spaces for your prompt (describing what you want to see in the image) and negative prompt (optional, describing what you don't want to see.) While you'll see a lot of examples of people making images with very long prompts, sometimes something short and simple can produce beautiful images. You can leave the negative prompt completely blank, and only add things if you see a need to.

The Prompt windows support some of the same shortcuts that you'd see in most local Stable Diffusion interfaces. You can select a word or phrase within a prompt, press Ctrl and the Up arrow at the same time, and watch that phrase get formatted as a higher priority part of the prompt. Don't go higher than about 1.4 with the weights. Anything above 1 is especially important already,

You can choose from three aspect ratios. You can make square images, which will be 1024x1024 pixel resolution if you are using an SDXL-based model. You can also choose Landscape, for a more wide-screen image, which will have a 1216x832 resolution, or Portrait (for a taller, more vertical image), which will have an 832x1216 resolution.  While none of them are fully wide-screen or as tall as a phone display, these are important controls that can have a big impact on your composition. Portrait aspect ratio, for example, can make more of a standing person visible in frame. (Your prompt will also influences your composition, of course. If you are getting shots that are too close-up on a person, mentioning their shoes or pants will also cause more of them to appear in the shot.)

At the bottom of the column is a yellow Generate button. The button displays the number of "Buzz" points that this generation will cost you. If you reduce the Quantity of images, you will use fewer credits. Under Advanced options, using fewer steps will also save generation time and use fewer points.

Generation of images seems somewhat slow, with images sometimes taking about a minute to generate. If you are switched to the "Generate Images" view, you will see an empty box or boxes with the word "Generating" until the image(s) are ready.

After images are generated, you can download them as .jpg files. In the 'three dots' menu on each generated image, Civitai lists Upscale as a "Coming Soon" option. This will be a very important missing feature, because the images generated are somewhat lacking in detail. If you have access to any leading Stable Diffusion interface that runs locally, you could bring these images in for a higher-resolution refinement, or at least upscale them and then do inpainting at a higher resolution to add details to faces and other key areas.

Give Civitai a try. I generated a lot of images for this review, and was able to do it all without paying for points. While I was working with all the models, I also found a new model that I liked enough to download. If you love Civitai and plan to keep making a lot more images, you can buy more points, or even subscribe to a plan for $10 per month. While the level of control and ultimate image quality might not match what an experienced user could do with a Stable Diffusion interface running at home, the selection of models and LoRAs is first-rate. Especially for people who don't have an AI-capable graphics card at home, this site makes it easy and fun to create a wide variety of images.