It can depend on a lot of factors. I used the lora generator available here.

For generation I set the steps to min 100 and try to set the batch size lower and number of repeats higher, depending on how much you want to pay.

I set the checkpoint to use when generating to pony. When autotagging I do 20 tags at 0.6 accuracy.

You preferably need at least 200 reference pictures, but it's not possible for every character. (I wouldn't go over 300)

Try to have pictures with only the character you're trying to recreate. (You can crop out other characters)

Use reference pics from different angles (front, back, side)

Try to have consistent art style for the pictures. Try not to have out of frame or too zoomed out screenshots.

Try to exclude weird expressions. If you want costumes to be accurate try to have a few screenshots showing the full costume, preferably from multiple angles,

When generating, if you didn't have enough reference pictures for your lora, you can try to use it at 0.8 or 0.6 strength and stabilize it with a style lora at 0.8 strength.

You can use this at the beginning of your prompt to fill in with a more high quality style:

score_7, score_7_up, score_9, score_9_up, BREAK,

Also use autism mix instead of pony.

My loras aren't the best, so I usually make characters I know will probably not get made.
You can probably get better advice from the more experienced creators on the site.

Good luck!