A quick tutorial on how I made the sample images for the "Where's the cat?" meme Lora that was created by aoaoao111111 (aka a111111111111111111) for my bounty. https://civitai.com/models/1051237/wheres-the-cat-meme-lora-for-ponyxl
Trigger word is " m3m3_cat ", useful tags are " english text, speech bubble, spoken question mark, ?, 3koma "
Putting 4koma and 2koma in negatives will also help
Aspect Ratio should be 2:1. (608x1216 for example)
"ForgeCouple" is the tool I used, and is the main reason the sample prompts look so complex.
I used ForgeCouple's Advanced Region Assignment to manually create the regions to keep to the comic format.
I personally use ' += ' as a separator, which divides the prompt into the different regions.
Here is the prompt from one of my samples:
<lora:cloudstrife-pdxl-nvwls-v1:0.75>, <lora:tifa-pdxl-nvwls-v1:0.8>, score_9, score_8_up, score_7_up, <lora:meme_cat_pony:0.9> 1boy, 1girl, m3m3_cat, english text, speech bubble, (3koma:1.6),+=
facing away, 1boy, defCloud, blonde hair, suspenders, shoulder armor, sleeveless turtleneck, open mouth,+=
1girl, facing another, defTif, black hair, low-tied long hair, earrings, white sports bra, black suspenders, black miniskirt, arm warmers, black elbow gloves, elbow pads, red gloves, nodding, closed eyes,+=
defCloud, blonde hair, suspenders, shoulder armor, sleeveless turtleneck, spoken question mark, confused,+=
defTif, black hair, low-tied long hair, earrings, white sports bra, black suspenders, black miniskirt, arm warmers, black elbow gloves, elbow pads, red gloves, (facing away, back turned:1.2), (closing door), doorway,+=
defCloud, blonde hair, suspenders, sleeveless turtleneck, (shaded face:1.2), wide-eyed, nervous sweat, scared,+=
defTif, red eyes, black hair, earring, (blush:1.2), facing to the side, parted lips, half-closed eyes, whispering, profile view, heavy breathing
The first line applies to the whole image, and the following lines I use to kind of "direct" the characters actions/state.
Here is what my ForgeCouple setup looks like for my sample images:
You can see how each line is linked by color to a region on the image, except for the first (red), since it applies to the whole image.
Even with all this, generations will still be a bit of a Gacha, since lines will sometimes get swapped around, but I'm still satisfied with the results.
If you've got some good generations, I'd love to see them.
If there's anymore questions, feel free to shoot me a message or post a comment, and I'll do my best to answer.