[Stable Diffusion] Prompt Sharing and Learning Thread

[Mr-Fox]

It feels like ComfyUI's ability to integrate two checkpoints and area control brings MidJourney's capability to our desktops.

Somehow it feels faster, like twice as fast as WebUI.

But! It feels like I still havent figured out how to do the "restore face".

Spoiler: Illustration

You don't have permission to view the spoiler content. Log in or register now.

By the way, it will break your prompts. Yet, I feel this whole pipeline concept is absolutely worth giving it a go:

View attachment 2468679

And I imagine the prompt is not immediately compatible:

Spoiler: Illustration

You don't have permission to view the spoiler content. Log in or register now.

Do you have to add all the nodes manually or is there template(s) ?
I have never been a fan of the node system because I don't find it intuitive. I'm sure it's just a question of getting used to it though. I can imagine that it brings some "new" opportunities to be creative and mix checkpoints etc. Perhaps it's more effective and less clunky in this way. My speculation only.

 

[Sepheyer]

Do you have to add all the nodes manually or is there template(s) ?
I have never been a fan of the node system because I don't find it intuitive. I'm sure it's just a question of getting used to it though. I can imagine that it brings some "new" opportunities to be creative and mix checkpoints etc. Perhaps it's more effective and less clunky in this way. My speculation only.

Every ComfyUI (CUI) png file is a node template. You drop it into CUI and it will rebuild whatever set up the other guy had.

I.e. I just dropped this official png

You must be registered to see the links

and I have this workflow:

 

[Sepheyer]

Mr-Fox here is one such "templates",

You must be registered to see the links

- even organized into groups. Load it using the file from the webpage:

Spoiler: Illustration

You don't have permission to view the spoiler content. Log in or register now.

And now this sucker is in my workspace:



And the image it generated using "clarity model" in my output folder:



FUCK. ME.

 

[Sepheyer]

Chonkers, The Next Generation

 

[Jimwalrus]

Mr-Fox here is one such "templates",

You must be registered to see the links

- even organized into groups. Load it using the file from the webpage:

Spoiler: Illustration

You don't have permission to view the spoiler content. Log in or register now.

And now this sucker is in my workspace:

View attachment 2468819

And the image it generated using "clarity model" in my output folder:

View attachment 2468822

FUCK. ME.

"Oi just want me nudie pictures of Taylor Swift, OK? None o' this nodes bizness!"

 

[Jimwalrus]

"Oi just want me nudie pictures of Taylor Swift, OK? None o' this nodes bizness!"

BTW, as it's now linked from the OP I've updated the Glossary

 

[Sepheyer]

CUI prompts are not usable by WUI and vice versa.

Vote with emojies:

Facepalm - new thread for CUI. GTFO. This is Automatic1111, the original post says so clearly. We can't use your stinking CUI prompts and thus this is a mere art sharing thread as far as the A1111 users are concerned and shouldn't be here. Just GTFO.

Waving hand - keep it here for now, no big deal really. The logic to keep both frontends together is to enable folks & lurkers to make up their mind thru osmosis whether they want to try that other frontend.

Vote count will be held on April 2nd.

 

[Synalon]

What Negative prompts do you use to stop text appearing in the image. I've tried no text, no letters, no words, and no tags but random words still appear when I try to generate pictures based on song lyrics.

 

[Sepheyer]

What Negative prompts do you use to stop text appearing in the image. I've tried no text, no letters, no words, and no tags but random words still appear when I try to generate pictures based on song lyrics.

I'd say the text seeps thru no matter what. Normally I have "watermark, graffiti, imprinting". Tho I think what matters is the style.

So, if you have "movie poster of a ..." you will have much more text than if you have "realistic photo of ...". With the poster you'll have looots of text, while with the photo you might have very rare watermark-like stamp in a corner that's easy to cure.

 

[Elefy]

no text, no letters, no words, and no tags

If you say NO in negative it meaning YES

 

[Synalon]

If you say NO in negative it meaning YES

Lol yeah you are right, I just have words, text, letters, poster, numbers, tags listed to prevent it, I didn't add the word "no".

 

[Sepheyer]

Sooo... ComfyUI doesn't have "restore face" feature. For the time being one can either use

You must be registered to see the links

or

You must be registered to see the links

or grab any other equivalent from GitHub. Naturally, rather messy.

Here we have before and after:

Original	CodeFormer	GFPGAN

The prompt is inside the original file.

 

[Mr-Fox]

Sooo... ComfyUI doesn't have "restore face" feature. For the time being one can either use

You must be registered to see the links

or

You must be registered to see the links

or grab any other equivalent from GitHub. Naturally, rather messy.

Here we have before and after:

Original	CodeFormer	GFPGAN
View attachment 2469376	View attachment 2469377	View attachment 2469396

The prompt is inside the original file.

Did you only use hi res for the last one? Have you found out your limit for the end resolution and/or the hi res multiplier?
I think GFPGAN looks the best at least in this comparison. I have found that it depends on the checkpoint and the image concept wich one looks best. In other words, one have to try out both every time before deciding.

 

[Sepheyer]

Did you only use hi res for the last one? Have you found out your limit for the end resolution and/or the hi res multiplier?
I think GFPGAN looks the best at least in this comparison. I have found that it depends on the checkpoint and the image concept wich one looks best. In other words, one have to try out both every time before deciding.

I was focusing on faces so far. What limit do you want me to try?

 

[Mr-Fox]

I was focusing on faces so far. What limit do you want me to try?

Oh I was talking about the resolution and the multiplier for hi res fix. The limit is when we get cuda memory error. I noticed that only the last one with GFPGAN was higher resolution. Was only curious.

 

[Sepheyer]

Oh I was talking about the resolution and the multiplier for hi res fix. The limit is when we get cuda memory error. I noticed that only the last one with GFPGAN was higher resolution. Was only curious.

Ah! I didn't test yet. The GFPGAN was upscaled by the Huggingface's online app. I am in the process of deploying the GFPGAN app on my PC, so I might end up doing my upscale workflow around batch upscale using the GFPGAN app.

Naturally, I'd stay entirely inside CUI, but its lack of "restore faces" is a bit of turn off. Though, I am still testing around, so my opinion might change within a few days

 

[Sepheyer]

This is how GFPGAN "desktop app" restored face of this chonker.



But for some reason instead of slapping the face back, it does this:

Spoiler: Illustration

You don't have permission to view the spoiler content. Log in or register now.

 

[Jimwalrus]

This is how GFPGAN "desktop app" restored face of this chonker.

View attachment 2469754

But for some reason instead of slapping the face back, it does this:

Spoiler: Illustration

You don't have permission to view the spoiler content. Log in or register now.

ABG Remover enabled?

 

[Sepheyer]

A complete roundtrip from CUI to GFPGAN to Krita. What a mess. Super high-touch:

Spoiler: Prompts

You don't have permission to view the spoiler content. Log in or register now.

 

[Sepheyer]

ABG Remover enabled?

The GFPGAN "desktop app" is a python command line tool, so imma be needing time to troubleshoot. It gave me a bunch of warnings when installing, so I guess the first order will be to reinstall it.