[Diskussion] Got questions about prompts and anything else you want to know, ask or share? Get in here!

[Stapfschuh]

right now, i'm downloading this checkpoint https://civitai.com/models/516638/feelingdiffusion?modelVersionId=574652
since i'm using a1111 and not comfyui, i couldn't test it. that checkpoint seems to run in a1111.

Post automatically merged: 15 Juni 2024

okay... that downloaded one also doesn't work. well, as i was waiting i read through some comments on youtube and my final thought on sd3 is that i don't need to download and test it due to its super heavy censoring. not the censoring in general is bugging me, he fact that you won't get any good image is it.

 

[Penniless7992]

"(((inverted-colors-style)))" comes out with some very interesting images. Thanks for finding another good prompt. I've been using "-art-style" for many months and now plan to experiment with mixing them together.

 

[Penniless7992]

Your horrible results in Pony based models is because you are skipping the "score_9, score_8_up, score_7_up" at the beginning of the prompt. It was an error in the original training that no one has been able to train back out yet. Also, you probably won't need so much prompt weight with parenthesis. Putting things like "source_pony, source_furry, source_anime, source_cartoon," in the negative helps with realism. Some models like "source_realism"/"source_real"/"source_realistic" but some turn it into an oil painting if those are in the positive. I'm not trying to be negative, but Pony models really are a different type of checkpoint. That is why loras and textual inversions made for other types of checkpoint go horribly wrong. It's like trying to prompt an English model with Chinese.

Prompt:
score_9, score_8_up, score_7_up,1girl, outside, relaxing,
Negative prompt: source_furry, source_pony, source_cartoon, painting, censored, 3d, cgi, blurry, watermark, simple background, bad hands, deformed,
Steps: 20, Sampler: DPM++ 2M SDE, Schedule type: Exponential, CFG scale: 7, Seed: 1808053711, Size: 896x1152, Model hash: 150fe1c0eb, Model: duchaitenPonyReal_ponyRealV10, VAE hash: 235745af8d, VAE: fixFP16ErrorsSDXLLowerMemoryUse_v10.safetensors, Clip skip: 2, Downcast alphas_cumprod: True, Version: v1.9.4


Image Width: 896 px
Image Height: 1152 px

 

[Stapfschuh]

i never can find any difference in quality between sd and xl. but pony just is hardcore when it comes to freedom and creativity.

 

[fullmanual]

Hello there. Longtime-follower-from-civitai-recently-found-this-forum. My question is about how to have a man and woman in the same shot, only the woman is wearing glasses. It seems like whatever is in positive prompt applies to everyone regardless - eyewear, hairstyles, race, etc. Would love to avoid regional prompting if possible. Thank you for still posting!

 

[Stapfschuh]

Hello there. Longtime-follower-from-civitai-recently-found-this-forum. My question is about how to have a man and woman in the same shot, only the woman is wearing glasses. It seems like whatever is in positive prompt applies to everyone regardless - eyewear, hairstyles, race, etc. Would love to avoid regional prompting if possible. Thank you for still posting!

right. this place called me a copycat. no way i would trust anyone from this place ever agian...

 

[fullmanual]

right. this place called me a copycat. no way i would trust anyone from this place ever agian...

Let me clarify. I'm not from civitai. That's where I found your article before I left.
civitai is ass-hoe for blocking some of my most innocuous content with no response.

 

[Stapfschuh]

Let me clarify. I'm not from civitai. That's where I found your article before I left.
civitai is ass-hoe for blocking some of my most innocuous content with no response.

the answer on what you are looking for is hidden inside my main thread. be patient and search for it. you will find a lot of other interesting stuff that will help you generate better images.