Search Results
7/2/2025, 12:00:01 AM
>>936514460
Play around on free sites like civitai and read the prompts of pictures you like. Note which model everything is based upon as not everything is cross compatible, particularly when it comes to incorporating low ranking adaptation files, or LORAs. These are specialized files for ensuring a more reliable output for niche visuals that can't otherwise be easily prompted. Also for specific character appearances if you're wanting to do art of others' IP.
Also my biggest tip is to use lora files to also learn useful prompts for things. Sometimes the trigger words themselves will give a good output from certain models without having to use the lora.
The rule of thumb I have is to build prompts from simple beginnings with lora setups I either know and trust, or without loras at all to see what I can do before I start adding too much. Too many loras/conflicting prompts/redundant and bloated prompts/poor prompt weighting can give you worse results.
And if the model just starts spitting out low quality crap, then do a throwaway purge image of some random and completely different prompt, then try again. Or if you're genning locally, just clear your vram, or reboot the whole pc if it's really stubborn.
Play around on free sites like civitai and read the prompts of pictures you like. Note which model everything is based upon as not everything is cross compatible, particularly when it comes to incorporating low ranking adaptation files, or LORAs. These are specialized files for ensuring a more reliable output for niche visuals that can't otherwise be easily prompted. Also for specific character appearances if you're wanting to do art of others' IP.
Also my biggest tip is to use lora files to also learn useful prompts for things. Sometimes the trigger words themselves will give a good output from certain models without having to use the lora.
The rule of thumb I have is to build prompts from simple beginnings with lora setups I either know and trust, or without loras at all to see what I can do before I start adding too much. Too many loras/conflicting prompts/redundant and bloated prompts/poor prompt weighting can give you worse results.
And if the model just starts spitting out low quality crap, then do a throwaway purge image of some random and completely different prompt, then try again. Or if you're genning locally, just clear your vram, or reboot the whole pc if it's really stubborn.
Page 1