Can confirm that in the latest ComfyUI, clip skip 1 results in a black image. Clip skip 2 works as expected.

@Cyberdelia No I use Draw Things. Clip skip 1 fails to generate. 2 works just fine though.

in fooocus also can't work with Clip skip 1 but fine with 2

in comfyui work on clip skip 2 not 1, 1 will result in black image

Forge is working with -1 while Comfy is not and requires -2

sd next also working perfectly with clip skip 1

@Cyberdelia What about to get rid of AI look once and for all?

@Cyberdelia Pony just knows concepts, what sdxl can dream about. Anyway much better anatomy, precise booru tags (you get, what you prompt), much much better emotions etc.

I switched back to xl from pony, because i was tired of the look and wanted some real photolike look, but yeah, people generate their dreams, which is easier with pony.

@gregoribanner970 i understand where you're coming from for certain concepts but i do believe the SDXL version is powerful with lighting options and more dynamic subjects.

@Cyberdelia like @gregoribanner970 stated it must be because of the booru tags and prompt adherence that Pony offers. do you recommend natural language prompts or tagging for this model? i tend to enjoy tags more than creating a paragraph.

@Cyberdelia oh that's interesting. how nsfw can it get?

@Cyberdelia thank you!

@Cyberdelia Do you have the system prompt or other info so that something like this can be run locally? That may be something you offer on patreon, I'm not sure.

@Cyberdelia Thanks for the response. I do have forge setup and running but I meant to ask if you had resources for your CyberRealisitic Prompt Helper LLM. I was wondering if you had available the system prompt, or any other information in order to set up my own local running LLM for that purpose.

@Cyberdelia Exactly that, sorry for the confusion. Thank you!

Popularity isn't necessarily due to the quality.

Pony can make a photorealistic image style, but the subjects are always stylized. Especially the faces are always wrong, people don't look like that.

im not. XL has a lot of issues that pony does not suffer from. while this is a good model, it is still XL.

yes

@Cyberdelia , In Comfyui do I use clipskip 2 or 1?

yes but you probably want pony for that. its better

@Cyberdelia yeahh

+1

Can't you just add the DMD2 lora to your workflow yourself?

@malari96 the generations don't come out anywhere near the same, when using a DMD2 lora compared to using a checkpoint with it integrated.

@TheSmizzle that's what i have been saying since the days of Turbo, then Lightning now DMD,creators don't try both and think we're lazy but difference is huge, without dmd generations lack details and need to finetune a lot while DMD integrated you get smooth crisp finish

Clip Skip 2 seems to make little or no difference in some models / prompts, but i sometimes get hugely distorted results without it. Like twisted limbs and wrong proportions.