Prompting Malifaux

IverCold asked in my previous post how I made chibi Malifaux characters.

It’s an iterative process. Usually I paste Wyrd’s official art into ChatGPT and ask for a chibi version, and I describe everything the character is wearing, their hairstyle and so on, as ChatGPT will usually fixate on one element.

The header image shows Rasputina dueling Perdita. Normally I would just render them separately, and then render the background separately, and composite them together in GIMP or Inkscape, but today I thought I’d try and do it all in one shot. It didn’t go well. ChatGPT really can only focus on one character at a time.

Perdita Ortega

I thought it would be fun to try rendering a non-Arcanist master.

This is Perdita Ortega, the master of the Family keyword, in the Guild faction. The Ortegas have become the Guild enforcers; they will go where they must to bring down their marks.

Raspy does not do well against the Family.

Prompt 1

Here’s the first prompt I tried for Perdita:

“Chibi anime-style young woman with long black hair with streaks of gray, wearing a very wide brimmed brown hat with a red hatband decorated with large silver studs, a tight low cut tank top, a brown leather bolero jacket, a long red sash around her waist, jeans with long, red-trimmed leather chaps, boots with spurs, two holsters, leather gloves, pointing a huge revolver, old west vibe. Expression: deadly calm and dangerous.”

ChatGPT rendered her with a tight black vest instead of her tank top.

Prompt 2

I pasted the official art for the second try, and added this:

“That is very close; I added the official art for context. Make her shirt a light gray tank top, as in the art. We’re looking for a female sexy but old gunslinger vibe. This is for the Malifaux master Perdita Ortega.”

Perdita Ortega official art

ChatGPT did a pretty good job, but used a cowboy hat instead of her iconic bolero hat.

Prompt 3

When making prompt changes, it’s always a good idea to keep including details. ChatGPT will not remember them if you don’t keep emphasizing them.

“rerender the image but make the hat a flat, wide brimmed bolero hat with a raised brim along the edge, brown to match her leather outfit, with a red hatband decorated with large silver studs.”

Close enough for government work. I just noticed that the official Perdita art has her left handed. Oh well.

8 thoughts on “Prompting Malifaux”

  1. Thanks for the post! I didn’t think that you begin with full description first. I tried “create an art of this character in cartoonish chibi style” with original art added. Then I added one of your Raspies as reference, then asked for “simplify details for a smaller version” and then asked for a pose change with reference.
    In the end I got this kitty Mei Feng:
    https://imgur.com/a/EC6kM6U

    • So cute! Mei Feng was my first master, and the one I was playing during the 4e beta. Still upset she left the Arcanists, and still even more upset that Foundry lost Sparks and Porkchop!

      I think she came out really well! MAYBE a little too much like M&SU Toni…

    • I rendered them individually in the poses I wanted them to have, specifying no background. I then generated several stage ideas in Midjourney, chose the one I liked the best, and composited them all together in either GIMP or Inkscape, can’t remember. Probably GIMP because I may have needed to remove a white background. Trying to generate more than a single character at a time just doesn’t work with ChatGPT; if you look at the header image, Perdita is WAY off model, and it had JUST generated the correct one. Raspy is also off model, but I was in a hurry. Also the background was not what I wanted; this is why I tend to use Midjourney. It does amazing backgrounds.

      • Thank you for the answer! I want to create a sticker pack for my local community (for one of the tournaments) and maybe (I am a mad man) a whole Chibbi themed fate deck (I love custom decks).

        • Oooh nice! That’s a great idea — I hope you’ll share it so I can see!

          I have a tournament coming up at the beginning of February — CaptainCon — I should do the same 🙂

  2. Chat GPT JUST released a new image model that is supposed to be much better at editing images. So in theory you could feed the image back in and say ‘change her hat to a bolero hat” and it would fix it.

    So they say. I haven’t had the bandwidth to try it out yet. (I think it came out yesterday?)

Comments are closed.