I’m entering a prompt and can never get an image close to what I ask for.

[jwstanley]

I am asking for a sloth sitting next to a skunk. But it keeps generating two sloths or animals that look like brown raccoons.

i asked ChatGPT for a prompt and this is the prompt it told me to use 

"Generate an image of a sloth and a skunk sitting next to each other. The sloth should have its characteristic slow-moving posture, with long claws and a smiling expression. The skunk should have its distinctive black and white fur pattern, with a fluffy tail and a curious expression. Both animals should be sitting peacefully side by side, with the sloth slightly leaning towards the skunk in a friendly manner."

however, I keep getting the same images

[Dave Gustafson]

Hi JW, I took on your problem, and it was a challenge lol. Everything you described arises in successive generations, a typical Firefly account would have burned through its tokens without any results. It seems this program has "characteristics bleed" which I have found in my work. As stated the sloth features bleed into the skunk, either creating another sloth or hybrid "brown racoons".  The engineers/coders need to take a look at this, it shouldn't be this frustrating to use if Adobe wants to be competitive and yes the program is still an infant. During my generations I submitted feedback concerning this "bleed." Anyway I brute forced it to "differentiate" between the two animals. 

This is the forced prompt: a happy "sloth" on the forest floor. ; with a large happy "black skunk" standing next to the "sloth."

This is how I got it to work in combination with generations and ratings/training. There is probably experts on here who would use "positional" prompts such as foreground, background, etc.., to seperate the subjects, but my efforts still had the bleed. Wish I could be more help. Cheers!

[Dave Gustafson]

I also have to add, stay away from chatGPT for your prompts. It's long-winded and bloated which adds another layer of confusion to Firefly in its current state lol! I'm still learning to write concise prompts myself, am I training Firefly or is it training me  *shrug* !!!

[Dave Gustafson]

I'm back because I just found this buried on an Adobe page.

"Other refinements include noting elements you’d like to avoid, specifying the strength of your style, and giving guidance to the AI in terms of how closely you want it to attend to your prompt. For example, if you want to save blue tones for a different character, just type [avoid = blue] at the end of your prompt."

Thinking this might work better than my "forced prompt". 

A happy sloth on the forest floor. ;with a happy skunk sitting next to the sloth [avoid=sloth]

If I'm reading it right, the sloth characteristics would be ignored, just like the characteristic color blue in their example. It might work or something similar. I would test it, but I need sleep lol. In closing I really wish all this "advanced" prompting was all collected and organized in an expansive comprehensive guide that we could easily reference, instead of "info mining" this on scattered pages.

[droopydog500]

I was also going to say that ChatGPT and Google Gemini generate prompts too long/too many filler words.  But they can give you ideas on some words and phrases to use.  When I wanted to reimagine what Whoville looks like from Horton Hears a Who, I asked ChatGPT and it came up with some phrases that generated some good results, like:

town with houses made of "shimmering squiggles" and "wiggly walls", streets are "twisty and twirly" and "up and downy", trees are "squirrely" and "wiggly", flowers are "twirly" and "swirly."

(This is not the exact original image because the styles were stripped from this favourite due to the December 1st Favourite bug.)

[droopydog500]

One other approach which might use less credits overall, is generate the two characters and the background independently, and the use Adobe Express (or Photoshop) to assemble them in one image.  That gets you around the characters "bleeding" into one another problem.