Intro

I’ve been seeing a rise in LoRAs with no or bad descriptions, which seems like a great way to get people to ignore your LoRA after you put work into it. I decided the best way to help fix this would be to write a guide on how to make a good one. I’ve already got a guide on LoRA making in general that has some notes on this as a section, but I think it may need a dedicated guide. You should consider typing up most of your description before your LoRA is even complete (use a local text document), as having an ordered and consistent tag list planned out will make tagging so much easier.

Before we start, one important note for typing on Civitai: Shift+enter makes a smaller line break than enter alone does. It's critical for making lists where each entry is its own line.

Things to include in your description

Who is your LoRa and What Does it Do

The most basic thing for a LoRA description, and title, to have is simply to say what it is, and yet I see so many fail to do even this. Most beginners will be doing a character, in which case this part is easy: "[Character Name] - [Media of Origin]" is a sufficient template for most character LoRA titles, while a description just needs to Include your character’s full name (for search purposes) and where they’re form (if the work has multiple incarnations, specify which). If you’ve got “special” versions included (for example, the default outfit and the outfit the character wore in a particular episode is trained), specify where it’s from and that it's included. Example:

Makes Jane “Janey” Doe, the 20 year old adventurer lead from the 1984 cartoon The Adventures of Jane Doe. Includes both her normal outfit, the party dress she wore in episode 10, and her swimsuit from episode 34 and 69.

It’s slightly trickier when you try something else besides a character, but the general idea is the same.

What Tags your LoRA uses

This one is pretty straightforward: List your trigger word, and words you trained the LoRA with. Without them, your LoRA is useless. Please list these in order of head to toe as best you can, as it will make it easier for everyone (you included) to exclude parts when adjusting the camera focus (e.g., full body, cowboy shot, upper body, portrait). For a character with multiple outfits, it’s best if you keep the character’s basic details on one line, then have each outfit or variant on a different line (you can actually do this even if you only trained a single outfit, since people like putting characters in different outfits). I recommend only including “1girl” and the like if it’s non-obvious, or it’s not 1boy/1girl (such as some kind of monster using 1other). I like starting each line with a name and colon in bold for quick visual identification.

Example:

Tags:
Base: JD_TAoJD, blonde hair, long hair, hair ornament, blue eyes, mole under eye, lipstick, large breasts
Main outfit: t-shirt, white shirt, print shirt, jeans, denim, blue pants, sneakers, black footwearDress: Black dress, strapless dress, side slit, flats
Bikini: bikini, blue bikini, strapless bikini, cleavage, side-tie bikini bottom, sandals

You can also list tags as “Negative:” if your training data has a large quantity of something people may not want in their generations (generally post SD1.5 models are pretty good at excluding things that aren't in the prompt, even if they were the majority of training data, if they were given a valid tag in training but it helps to have the option of throwing them in negative too if needed) or fixes to issues you see pop up in generation (Is this article of clothing commonly misdrawn as a different kind of clothing? Use a negative for the kind it's not supposed to be.). If you have something with extant but limited data (such as some article of clothing you’ve only got a few pics for in data), note it with its limitations: It will sometimes work better than expected.

These tags (except negatives) should also be written in the version details (the step after the description) by unselecting “This version doesn't require any trigger words” and including them. This form supports keeping words together in “blocks” by writing/pasting the grouped words together at once, separated by commas (,) before hitting enter (or clicking the thing that starts with “Add”). As in the description proper, I recommend keeping the character’s base attributes in one block, then each outfit in its own block

If your LoRA doesn’t have a trigger word (really only appropriate for a style LoRA), just say none is needed.

What was your LoRA trained on (optional)

A short count of your training data and what kinds of pictures are in it. This one is optional, but it’s really nice to have for helping understand just what kind of results your LoRA will make, and helps future creators (how many images you need continues to be one of the great debates of LoRA making).

Example:

Trained on 69 screenshots of the cartoon (21 in her main outfit, 15 in her dress, 16 in her bikini, and the rest close ups of her head), and 3 pieces of promo art (all main outfit).

Populating a gallery

Nothing says a creator put no effort into a LoRA (and thus it’s likely not worth using) than a gallery consisting of only low quality epoch pics long after posting. This is easy to fix though. After posting a LoRA, you can edit the gallery by clicking the three vertical dot icon and clicking edit post. This allows you to add new pictures to the gallery and rearrange it (You can keep your epoch pics, but I’d recommend sticking them in the rear of the gallery so they aren’t the face of your LoRA). The cover pic of a LoRA, the one people see when your LoRA is in the list of LoRAs, is the first image in the main gallery with a low enough age rating for the user to see (this means you need at least one PG image to be seen by users who aren’t logged in or are on Civitai Green), and you can change which one is first by using the Rearrange option in the upper right of the post management.

For a character, I recommend your gallery start with simple pictures that showcase each form you've trained and using as few other LoRAs as possible (especially avoiding art styles unless you really need it). Just some standing full body, cowboy shot, or upper body pics of your character in each outfit you support on a basic background (but preferably not “simple background, white background” for the cover pic. “grass, plain” or “floor, wall” is enough), and possibly some pictures from behind. After that minimum is accomplished you can start showing off crazier stuff: Yes, if your character is an adult woman showing you can take her clothes off is nice, but unless you’re making an SD1.5 model, that’s not actually impressive and using it as your cover image makes your model blur into all the other many pictures of naked people on the site.

Writing a good description for your LoRA