🏞｜general-with-images

[vast wedge]

94-29-09-786-5-13

[tame wharf]

did u use comfyui and savro??

i like the lighting effectss

[vast wedge]

I used image generator pro

[tame wharf]

oh coool

[vast wedge]

Yeeaa getting it to be exact lighting, I prompt it” I will give you another photo use its (aesthetics,tone,color,etc.)” to make the picture I’m generating follow a script

[ashen yoke]

Epstein survivor Juliette Bryant says he shape shifted into this, how would I prompt to get something similar

[analog ruin]

[[maybe try this prompt]]] "A highly detailed portrait of a male lizard-man resembling an Argonian mixed with an alien, with dark fantasy influence. He has vivid green, slightly scaled skin with a humanoid facial structure. His eyes are large, deep black, and demonic, reflecting no light, giving a haunting expression. A long, red, forked tongue extends from his mouth. Organic frills, like leafy or flame-like protrusions, rise from his head, giving him a crowned, dragon-like silhouette. The expression is stern and intense, the lighting moody and cinematic, with soft shadows. The setting is neutral or misty, to emphasize the creature's face and upper body. Ultra-realistic rendering, hyper-detailed fantasy art, concept art quality, 8K, by Greg Rutkowski and H.R. Giger, dark and mystical atmosphere." [[[and this is the negative promt]]] "cartoon, anime, smiling, goofy, friendly, low-res, low detail, overly scaled, open mouth with teeth, bright eyes, robotic, glowing eyes, childlike, soft shading, extra limbs, blurry, humanoid with human skin, horns, Armor, fur, metallic, happy tone, stylized game art" (idk im a noob here lol)

[rich sun]

cool. i like the bnw effects to the image ❤️

what are u using @manic stone

[balmy reef]

love the cool mix of futuristic tech and mythical phoenix imagery mannn,,,cool!

[solar tide]

is it better to just openpose manually than using the one in controlnet, idk if its my settings but it doesnt include hands and feet overall seems a bit weird

[hearty violet]

guys, can someone give me tips about upscale? I did this image but I dont like the result

I don't know, I thinking I'm doing something wrong

how you do it?

I use controlnet tile with sd utlimatium upscale

[hearty violet]

you are right, this image looks good! But I generate a lot of then and like onyl this one, I'm in doubt about my methods. I will check on png info how I do it... Can you share about your process, pls?

the goal is add more detail

sure, thanks! 🙂

the first one was very good, how you did it?

multdiffusion?

my god, only it? 😮

an d your prompt?

what you means for each section? is not a prompt for all the image? You tiled it?

ok, I will try

I will put on gemini and ask to make a prompt for what it see, and use to uspcale with ultimatium 2x with the same prompt

ok, thanks a lot! : )

[solar tide]

Hello, guys, I have a question. Is there a way to get Clip Skip 2 in SDXL Forge without going to the 'All' UI in forge? In A1111, I just changed it in settings, but I don't think that setting exists in Forge.
The reason is, when I go to the 'All' UI and generate an image with the same settings and prompts I used to generate an image in SDXL, the results and style vary so much. So if possible, I don't wanna use the 'All' UI and I want to just change Clip Skip 2 in the XL UI without switching. From left to right, starting from the blue background image: XL UI unknown clip skip, all UI clip skip 1, all UI clip skip 2

as you can see, the XL UI is just better, so I don't wanna use the all UI

[hearty violet]

@manic stone is this configs right on ultimatium upscale?

I change only the ultimatium, I put the denoise on .12 as you said

what? how can I do it?

"steps" where?

to my maximum resolution?

I will put the higher value so

my god it works

I'm generating the second upscale, I will show you : )

you means changing the upscale model?

instead 4x real upscale tryu this siax?

nice, thanks! I will take a look on it

I like this one
https://openmodeldb.info/models/4x-RealWebPhoto-v4-dat2

left (before) / right (after)

you help me a lot, thanks

I apreciate your help, gentleman

https://tenor.com/view/winnie-the-pooh-gentleman-gif-11290485799815337210

[wispy nest]

if anyone has experience lora training ive been trying to train a lora to produce images in yusuke muratas art style from opm. ive been running into some hiccups post creating the lora that id like some help understanding

[solar tide]

does anyone know how to get hands and feet to be apart of their openpose map ? right now im using openposesea website and controlnet to generate an openpose but they only include the main body frame and not the hands. the displayed image from openposesea and the exported image when i download it is different.

[arctic wave]

loop

[wheat crescent]

https://www.soulify.space/

[hearty violet]

@manic stone I forgot to ask you, I must use the same seed in the upscale or not?

[solar tide]

would someone be able to help me out by generating an image with controlnet using an openpose reference ? cause im getting very weird results everytime i use openpose. id like to know if the problem is in my settings or if its in the reference image itself.

its suppose to be a guy doing pushups and ive had no problems with depth and canny but openpose gave weird results

[raw vector]

@deft leaf @errant hornet @reef olive not ded

fucking discord/gmail is total ass

and i think i may be permanently locked out of my old account

[deft leaf]

what have they done to you?! good to hear signs of life though!

[raw vector]

idk but i just found the L3 invite so i'm back over there now

[dusty sapphire]

#🏞｜general-with-images make a chair

[rich sun]

looook guys

[last moon]

hello guys, can someone help me, i wanna do twitch emotes for my channel with chibi-Sylvana, somehow i make good one emote, but i need to make couple different ones with different expressions, but my problem is that i dont know how to make another emotes with same face, colors and armor. When i try generate new images with different exressions - armor, color or face is changing. 2 Examples:

[wheat crescent]

https://youtube.com/shorts/c3qW1Bn4y9w?si=EL-oQsBMXJSe_jVZ

[last moon]

Can you please explain in more detail? I am a complete newbie

I make my emotes local too, with Stable Diffusion, if that's what you mean

yeah thx, i just download ComfyUi, do i need some special settings to do it?

thx bro, i'll try it. Can i tag u if a have questions?

ofc bro, i just dont wanna bother u

im just new in this AI things and first time using Stable Diffusion and similar things

[last moon]

bro, for what i need this

oh, ok, thx

when i installed this bitch-program, she suggest install this files, idk why my folder is weight 500mb when i downloaded 15 gb files, or im stupid idk

[last moon]

thank u so much,bro, after fixing some problems, everything worked. I'll try to experiment with this.

❤️ and sorry for my english)

[solar tide]

@quiet current here is an example. there is similarities in that the guy is lying down but its not a push up

[quiet current]

you're probably not setting it up correctly

blind shot that turns out right 9/10 times, you're not using the correct controlnet model. Aka you're trying to mix a SDXL model with a sd1.5 controlnet model for instance.

or sd3, sd2, whatever combination, instead of sticking to the same architecture all across your pipeline.

[solar tide]

hmm ok then i will double check my settings. thank you 😄

[rich sun]

cooool

[rain gazelle]

I like this one. Has a renaissance oil feel to it.

[magic oak]

Path of the king, my latest artwork

[clever hull]

I have a question. What could be the possible reasons for the slow loading of "manage"?

[celest sigil]

Have you only just started ComfyUI? Is it still doing this in the console, or has it reached "DONE"?

[wooden sand]

expand this

[odd dune]

Here’s a cinematic character face pack I created with Sora and refined in Photoshop.
10 ultra-high resolution portraits with detailed expressions and AI-enhanced textures.
All visuals come with a mini lore sheet in PDF.
📦 Full pack available here: https://payhip.com/b/AwaXc

[fleet goblet]

I'd be amazed if anyone would pay for this.

[hybrid prism]

Why don't you join our agency to make money?

In that case your income will be much better than this image selling work.

Without that much effort.

[fleet goblet]

^ Staff? We got shills in the general images channel...

[hybrid prism]

Unfortunately the owner. Lol.

[fleet goblet]

@wheat girder We got weirdos advertising their "Agencies" and selling images in here.

[hybrid prism]

OMG.

Never advertising.

[fleet goblet]

Nah, you are part of this.

[hybrid prism]

Just giving someone advice.

I don't know him. Dont make me laugh. 🙂

[fleet goblet]

You clearly are a bot or something.

You don't even type that well.

Your account's also new.

[hybrid prism]

You saw the bot that types in the wrong way?

[fleet goblet]

Not hard to make a bot mistype. Systemprompts are a thing.

[hybrid prism]

😉

Brilliant!

But not good at finding real bots.

[fleet goblet]

just gonna block you and leave your talking to yourself

[hybrid prism]

Maybe you could be a bot.

You are the bot?

Ur so gullible.

Anyway, nice talking to you @fleet goblet

[brazen schooner]

I've been trying to make a more controllable local workflow similar to the meshy product using SDXL as base for image gen. (The idea being that the base image is always projected, I'm still working on automating that process) But getting closer to results I like running locally on an a 1080ti.

[wispy nest]

hello yall. ive been working on making a opm/murata style lora. these 3 images are 3 different models, just wondering what yall think

[rich sun]

i like it

[rich sun]

coool stuff

[rich sun]

heyyy. can u recommend me ai tool where i can enhance this? i made this on savro. id be glad for any suggestions

[tame wharf]

made this on savro, tips how to enhance this result or another ai tool

[silver moth]

give me a image of a terminator like humanoid

[raven mantle]

Hello

Make this a 4k edit

[woven pollen]

@hybrid prism my latest pic I generated, tried to generate smth fitting the theme
'Togetherness in winter wonderland'

[hybrid prism]

Awesome!

THe content is very unique.

You are a kinda writer?

[woven pollen]

I´m generating pics with SD 1.5 and Perchance, write Songs and bring them to live with Suno, wrote a 52 pages Guide on how to use Perchance and write a Fantasy Book with the help of Gemini

I mostly generate landscapes and SFW portraits of females

Perchance is a website where you can generate pics on for free (SFW / NSFW) and its based on SD

[hybrid prism]

Beautiful!

You are using SD1.5 locally or just the online service?

[hybrid prism]

Could you please let me see that guide?

[woven pollen]

I have some other pics I generated to that theme, if you wanna see them

[woven pollen]

sure, would dm be fine?

[woven pollen]

locally

[hybrid prism]

Of course.

[hybrid prism]

Please.

[mighty hill]

hello. is there anyone who have ideas about base model, loras and extensions might be used generating this one? pony as model and one of loras is floox style?

[silver moth]

hello, is there anyone who have ideas about base model, loras and extensions might be used generating this one? pony as model and one of loras is floox style?

[balmy reef]

u made that with savro? maybe u can enhance it with comfyui man haha

[marble pewter]

Free TradingView Premium, Full Version (Windows & macOS):
https://www.reddit.com/r/CryptoForexSyndicate/comments/1kxeejv/

[onyx halo]

guidance == cfg scale?

any other params?

I will look into it when I reach home downloading the model on cellular is horrendous

[onyx halo]

update @manic stone flux1_kontext def does not wanna load under sd.cpp, dont exactly know why but i opened an issue at their github, now we waut

wait*

[onyx halo]

the descent into insanity continues

[onyx halo]

Flux runs at 27s/it jfc

[onyx halo]

with sheer willpower yeah that's what it's running off of at this point

Problem is it doesn't do what I want but I think that's because I am missing a few params

sd.cpp resizes my image to 512x so I am assuming that's where it loses the data to work with

No like

My original image isn't a proper multiple of 64 in dimensions

just fixed that and hoping it works now that I have specified width and height too

so it doesn't auto resize before using the image

If this doesn't work I WILL fold at this point

[onyx halo]

wait

cfg 1

[DEBUG] stable-diffusion.cpp:1182 - computing vae [mode: DECODE] graph completed, taking 61.39s
[INFO ] stable-diffusion.cpp:1827 - latent 1 decoded, taking 61.39s
[INFO ] stable-diffusion.cpp:1831 - decode_first_stage completed, taking 61.39s
[INFO ] stable-diffusion.cpp:2079 - generate_image completed in 1561.15s
save result PNG image to './her.png'

I WAS RUNNING IT ON CFG 5-10

[onyx halo]

@manic stone eyo wanna hop over DMs rq?

[mighty hill]

it was supposed to be funny I guess

[hearty violet]

https://www.youtube.com/watch?v=C53KEXOnZks

the video is already translated. It can have some mistakes, I translated the text with canva and the audio from pt to english, but if I do it I will pay a professional 🙂

[dawn fulcrum]

cat

[woven pollen]

@spiral gate I was generating pics like that

and thats the pic I could imagine on a mousepad

[spiral gate]

ah yeah, this is pretty sick. I see how it would fit

but don't anime fans like yourself hate AI?

[woven pollen]

Hate it?
I'm a big fan of it 😄
I even gonna work with it in my job that I start next month

I use it to work on pics, music, books and generally in my daily life

[spiral gate]

cool, whats your job?

but what I meant was, doesn't the majority of the anime community hate AI?

[woven pollen]

IT mixed with accounting

[woven pollen]

Mh, to be honest, I don't know

I'm an anime fan and I like AI
I don't know too many other anime fans, only like 4 or 5 and they also like AI, as much as I know

[spiral gate]

ah okay then, that's interesting

[solar tide]

hey guys working on making consistent characters with consistent accessories, like belt, necklace etc while in a dynamic/action pose. wanted to share my before and after pics and have a question ! (the original generated image lacked the green belt, consistent boots, green necklace, brown gloves, nice buttons)

these were my best tries but i relied on just randomly generating till i got the output i wanted which seems unreliable and the generation itself isnt that consistent to the lora. i used controlnet depth + canny when inpainting/inpaint sketching too (the lora i used: https://civitai.com/models/670113/guy-cecil-tales-of-the-abyss-pony-illustrious)

[mighty wren]

I will be grateful if you liked my artwork:
https://sora.chatgpt.com/g/gen_01k13xvwemfectqt3yf4wacwgm

[jovial solar]

Does anyone know what's up with the ai hate? Like I understand that it can be tiring seeing many low-quality images, but recently the quality went up drastically, and many generations look really peak? It gets to the point of me specifically pointing out that I don't compete with artists and that I understand the fact that what I do is different, but still get a ton of hate—it's frustrating -_-

[edgy ravine]

@vast flume

[halcyon burrow]

Ikr! I'm constantly getting hate from people over Ai, it's frustrating. I usually hear two main arguments from them though.

1 - Ai is bad for the environment (Requires massive server banks filled with millions of GPUs that draw insane amounts of electricity) Which I guess is true, but not for those of us just doing generations on our at home computers. Also, it's common for new innovations to initially be kinda crazy like that, but you have to start somewhere, and in time as we get better and better at training AI and making better Ai it will take less and less to run them, besides, the improvements new innovations bring to humanity is worth it imo. I think it's a silly argument, like, dude, if you don't like modern advancements that cause pollution, then go back to riding in horse and buggies and sell your car.

2 - Many Ai art generators are trained on other peoples art without their consent, royalties, or even acknowledgement of the original artists. So they say that Ai art is unethical. But to that argument I say that humans do the exact same thing and always have. Any time a human artist creates an image, a painting for example, they are pulling from past life experience, from every image they have ever seen, every tv show they have ever watched, every other artwork they have ever looked at and using all of that as a reference point to make their own art.

Anyway, rant over. But yeah, the hate I feel like is unwarranted. Ai is awesome.

[restive trench]

@halcyon burrow I've experienced vitriol reactions as well.
1 - Ai is bad for the environment.
So is oil, logging, cattle farming...etc..
2 - Many Ai art generators are trained on other peoples art without their consent
Gotta quote Salvador Dali on that one... "Those who do not want to imitate anything produce nothing."

[halcyon burrow]

That's a great quote, love it.

But yeah, I play a lot of table top games over the internet and sometimes I take it upon myself to "upgrade" someone's character art, of my own free will, thinking that they will be impressed, grateful, and amazed, and 90% of the time they are not... They are actually upset, angry, and tell me to "never do that again" lol.. I just take their image and do a general upscaling, keeping the denoising low to preserve the integrity of the original image while uping the cinamatic vibe of the art, makig it have better lighting, contrast, shadows, and more detail, and they get upset.

[jovial solar]

I've been doing that too a while ago, and ppl have done it to my work, and what I can tell is that it can do some serious damage to one's pride😅 It would be a rare find to come across a person who's so secure about their abilities that they'll appreciate that, sadly
Also, continuing the message I sent earlier:

1. Posted some character images that I consider good
2. Added the "Ai-art" flair and spoilered the entire post
3. Asked for some opinions, also asked to refrain from hollow AI flaming
4. The first guy immediately starts being rude, also used "we" and "us" instead of "I" and "me", although I didn't notice that at first
5. I retorted a bit, and than asked him about why he thinks that AI always equals "shit", but he didn't respond, lol
6. My post and comments get disliked into oblivion
7. (Back to number 4) Checked the guy's profile -> "Oh, he's a part of the artists community, I get it now"
8. We're back to artists flaming any AI user and not having the smallest fraction of objectivity while trying to get their point across. Now I'm genuinely convinced that there's a specific kind of people who don't post art themselves and instead repost all art that they find, and in the same time are babies without the ability to coherently express themselves, who also start throwing a tantrum the moment they're upset

Like at first, I tried to get into a civilized discussion, but now I kinda do want the smoke from them😭

I hate irrational people so much🤌 Maybe I'm in the wrong in this situation, but I'll never f*cking know if the other side literally returns to monke🤷‍♂️

Since we're doing quotes, I'll throw one in to lighten the mood a bit : D It's about emotion, so it fits the theme too
"If darkness chooses shine, it weaves a night of stars
If tears embrace that shine, they dance like falling stars" - Your lie in April, "Hikaru nara"

[solar tide]

can anyone help me fix the shadows? i tried inpainting but does not work... and could not find a shadow lora

[dusky marsh]

hey guys i keep having an issue where my stable diffuson makes multiple faces in one image

prompt:
A photorealistic close-up solo portrait of one beautiful young Caucasian woman, headshot, looking to the side, soft lighting, smooth skin, long flowing hair, gentle expression, shallow depth of field, soft bokeh background, high detail, natural makeup, cinematic mood, 8k resolution, DSLR photo

negative Prompt:
multiple people, group, crowd, two girls, three girls, extra persons, blurry, lowres, bad anatomy, distorted, cartoon, unrealistic

samp steps 30
resolution full hf 1920x1080 cfg scale 9

samp method dpm++2M

[halcyon burrow]

That happens to me a lot because of high resolutions. If I try anything above 1024x1024 resolution, I get multiple faces. If I want something higher than 1024x1024 I have to use hires to upscale x2 or img2img to upscale x2 and have to keep the denoising to .4 or lower while upscale or else again, multiple faces and other issues.

That might be your problem? I am no expert.

[dusky marsh]

ill try now with 1024 thanks, also what sampeling method is best for realistic humans?

[halcyon burrow]

I don't know the answer to that question, sorry >_<

[dusky marsh]

thanks anyway, ❤️

is there a way to also make nsfw with sd?

[halcyon burrow]

Did that fix your multiple face problem?

[dusky marsh]

i still get some multiple faces

even tho i specify

[vagrant dust]

Check the version of model you are using. SD 1.5 Models are 512x512 base resolution. SDXL an most other current use 1024x1024 as base. You can chose portrait or landscape format if you keep next to the resolution. For example 896x1168 etc.

[vagrant dust]

I would say dpm2m and karras but everyone got favorite combinations of schedulers and samplers.

[dusky marsh]

i will try that, thank you so much

[vagrant dust]

Happy to help, here is your shadow free night image....

[dusky marsh]

❤️

[solar tide]

@vagrant dust now thats what i call thinking out of the box haha

[vagrant dust]

It seems that flux kontext don't want to change the shadow. Not sure if it understand the concept.

Changing the shadow into a shadow of an airplane works .... so well....

[solar tide]

@vagrant dust its all good ill try manually drawing it then refine it with inpaint maybe thatll work. thanks for trying bro !

[plucky hamlet]

Inpainting should work. Draw the shadow in the same color as the existing shadow and remove shadow with the grass color. Then mask only the part that needs to be redrawn and use flux fill. Play with the denoise.

[dry blaze]

People may hate on vibecoding, but i be here un-deprecating diffusion pipe

The error i'm attempting to have fixed to be less deprecated lol

[drowsy knot]

Hello @dusky marsh
I also experienced this same issue yesterday. Regrettably, I can't share linked articles here, but feel free to hit me up—I’d be happy to share some helpful resources with you!

[wispy nest]

Middle Eastern-inspired Sub-Zero and scorpion.

[shrewd falcon]

Anemone (Kizetsu Yuusha) V1 illustrious Apex_Ai

[dry blaze]

Alrighty, torture test runs for my new multithreaded python hub project TLDR,

give it the path of comfyui, and it will auto install it, and any custom nodes's requirements.txt it finds on boot,
auto detect gpu and drivers,
auto download compatible torch with cuda if you got nvidia,
auto detect which gpu you have, which cuda version you got in path to cross reference there,
if your gpu driver is outdated, it will tell you and link to where you download them,
python hardlink files for massive space savings if you got multitude of python programs that has their own venvs.
if it detects you added a custom node that is incompatible with the rest, it will ask you if you want to remove that custom node, or automatically create a new comfyui just for that custom node, stableaudiosampler for instance, or comfyui AIT which hasn't been working for months, or almost years for me lol.

And yes, the bloops you see in the image is what i'm having it fix now.

[dry blaze]

Damn, nice! Sadly i have no idea how to make sales of anything lol

[grim narwhal]

Com certeza! Aqui está o prompt traduzido para o inglês:
"A captivating and persuasive book cover design for 'A Recompensa da Honra'. The artwork should visually represent the journey from humility to honor, illustrating the principle that humility precedes honor and that honor opens doors. Portray a serene, humble figure or a subtle, ancient-looking pathway emerging from a shadowed, quiet space into a brightly illuminated, expansive vista or a grand and open architectural entrance. The style should subtly blend elements reminiscent of early Christian aesthetics, such as soft, ethereal light and timeless, simple structures, with a contemporary touch, avoiding explicit religious symbols like a cross. The overall composition should evoke a sense of peace, elevation, and new opportunities, inviting readers to explore the profound meaning of honor. The title 'A Recompensa da Honra' should be prominently displayed in Portuguese."

/Com certeza! Aqui está o prompt traduzido para o inglês:
"A captivating and persuasive book cover design for 'A Recompensa da Honra'. The artwork should visually represent the journey from humility to honor, illustrating the principle that humility precedes honor and that honor opens doors. Portray a serene, humble figure or a subtle, ancient-looking pathway emerging from a shadowed, quiet space into a brightly illuminated, expansive vista or a grand and open architectural entrance. The style should subtly blend elements reminiscent of early Christian aesthetics, such as soft, ethereal light and timeless, simple structures, with a contemporary touch, avoiding explicit religious symbols like a cross. The overall composition should evoke a sense of peace, elevation, and new opportunities, inviting readers to explore the profound meaning of honor. The title 'A Recompensa da Honra' should be prominently displayed in Portuguese."

[hearty violet]

guys, for this example

why with depth I got this kind of result

and with canny the result is very worst?

canny turn the image so badly, why? It don't have onyl to mapping the lines? It change the quality of image?

[warped zenith]

Hi, I haven't used A1111 for a year. Today I realized that Segment Anything is down here and looks small. I used to use it in a tab in a large window; it was more convenient to use. Can someone help me? (old sd 1.5

[shrewd falcon]

Sui-feng / Soifon (Bleach) illustrious Apex_Ai

[halcyon burrow]

Lol... there's so many weird scams on this discord.

[dry blaze]

Auto is shit. It's last update was almost a year ago. Stick with forge or comfyui

[dry blaze]

Yep. Sadly fuckall steps taken to prevent them/auto delete them by discord

@wheat girder Need some cleanup. Copy their ID and search the server, as the 2 compromised accounts above spammed the scam in every channel.

[young escarp]

Hello! i wanted to share the images i just updated to my website, its all made with Stable Diffusion 3.5L , AI portraits ,zombies, sets and vehicules + 3 short films , thanks for watching ! https://www.brunohajnal.com/

[wide breach]

I'm trying to create an image of a demon and three human people but it keeps making every look like the demon. How can I get the 3 people to not look demonic please lol

[restive trench]

Look into regional prompting. You may need to run a sampler against each masked face.

[hearty violet]

I will try sd3.5 for the first time, how work the controlnet on this?

I found this page, there are more?

[ripe cedar]

What's a good way to make something realistic but with good coherence? I want to make a boxer stepping on a defeated opponent in the midst of rubble outdoors

[hearty violet]

where I got this VAE / Text encoder to Forge/ Reforge? I don't found it on settings to load

[lethal basalt]

Just a quick note—if anyone ever needs help with automation (bots, scripts, alerts, etc.), I’ve got some experience with crypto-related tools and I’m happy to help out. Feel free to check my profile or reach out anytime. No pressure—just putting it out there.I’ve found it useful, but as always, I recommend testing things yourself and doing your own research before fully committing."

[shut panther]

Hey all, I'm new here but I wanted to share a sick chrome extension I just made. I'm working on a mobile game that requires lot of characters, so i built a prompt generator.

[brazen sand]

but at the same time, what was the great breakthroughs after Automatic1111? I was wondering that, it seems that for generating locally there is no great advance other than you could achieve in Automatic1111.

Also for working in images it was a beast. Comfy seems so hard to set up and achieve same level of usage to me.

For example I did some 3D depth map to image and it worked really well, and you could keep going adding detail with inpainting etc, where in Comfy all proccess seems to need a giant nuddle of nodes.

At some point after the SD 1.5 to SD xl I thought image generation would keep developing, I don't know what are the advances right now.

I'm using Sora right now, it works just like I would want SD to work, but well.

But I'm aware that Auto is what it is now, maybe I'll give Comfy a try, but all examples are just completle processes that takes ages and need tons of specific models to work and usually they just don't.

Automatic was just so well made

and things haven't go really ahead, I would still inpaint with Automatic and such

[dry blaze]

Aye. That's why i mentioned forge. It basically took over the torch for automatic1111 and fixed the annoyances that auto struggled with. for instance a big bug auto had was break generations when you model hopped. (generated a image, switched base model, generated some more, or when you made a checkpoint XY grid

[elder talon]

Make this girl a super model, make beautiful long hair and bright makeup. Let her be in a long chic diamond dress and beautiful stylish shoes. Make her a beautiful background in the castle

Make this girl a super model, make beautiful long hair and bright makeup. Let her be in a long chic diamond dress and beautiful stylish shoes. Make her a beautiful background in the castle

[oak apex]

A dynamic vertical split-scene poster featuring a professional wrestler. At the top of the image, he stands dramatically on top of a ladder inside a wrestling ring, arms outstretched, facing a roaring crowd. He has long black hair, turquoise tribal face paint, and wears red elbow pads. Below, in the same image, he is flying mid-air in a Swanton Bomb move, leaping from the ladder toward an unseen opponent. The crowd and arena are in grayscale, while the wrestler is in vivid, high-contrast color. Comic-book style, intense motion blur, dramatic lighting, full body view –v 5 –ar 2:3 –style raw

[wispy nest]

Girl walk

Porn

[quiet current]

Really....

[still rose]

that was also my reaction

[cedar solar]

Hi everyone 👋
I’ve recently started creating stories and visual moments with Stable Diffusion, and I’m really inspired by how AI can bring emotions and atmosphere to life. 🌌✨

One of my first full works is the "Veranda Moments" series – a short visual story told in images, where mood and subtle details guide the imagination.

I’d love to share more with those who enjoy narrative-driven AI art.
If you’re curious, you can find me here:
📸 Instagram: https://www.instagram.com/nyxmoments.official/
📖 Full stories & packs: https://nyxmoments.gumroad.com/

I’d also be happy to connect with other creators, exchange ideas, or just get feedback. Thank you for letting me share, and have a wonderful creative day! 🙏✨

[daring kraken]

Keep the original photo exactly as it is — same person, pose, clothing, facial features, body, and background. Add realistic, glowing angel wings to the person, naturally attached to their back and softly lit by the streetlights and moon. The wings should blend seamlessly into the lighting and shadows of the scene, with soft feather textures and light effects. Do not alter the background — just enhance it by adding a realistic starry sky, a softly glowing moon, and subtle lens flares or atmospheric glow. Add a faint ethereal aura or soft light surrounding the figure for a divine feel. The final result must look like a real, cinematic night photograph, not AI-generated or illustrated — just a magical moment captured by camera.

/img2img image:prompt:"add glowing angel wings, starry sky, cinematic night lighting, soft light aura, photorealistic" denoising_strength:0.4

[high vortex]

installed stable diffusion from a tutorial but image is always a black screen. chatgpt was no help

[void shale]

you use comfyui

instead use it by downloading in terminal locally maybe it will helps you

[viral frost]

It looks like you are using a second copy of SD 3.5 L as a text encoder (sd3.5_large_fp8_scaled.safetensors). The third text encoder should be T5 XXL. Also, SD 3.5 Large is, well, large. You may not have enough vram to run it. Have you looked at the terminal output to see if there are any error messages?

[dawn seal]

She's already beautiful!

[cedar solar]

"Two worlds. Two temptations.
Which one would you surrender to?
✨ Choose your side on Instagram:
https://www.instagram.com/nyxmoments.official/

[viral frost]

I've tried out the new Flux.1 Krea dev and have mixed feelings. On one hand, it produces much more varied output, with more esthetic differences from image to image, and seems to follow prompts a bit better. But on the other hand its attention to detail seems to have been ruined. The things I most appreciate about Flux dev are its pixel-level detail and the general coherence of those details. SD3 always has a general blurriness and smeary, indistinct details, with edges morphing into each other and it being difficult to tell what small objects are. Flux dev makes individual shapes much more distinct as physical objects. Krea looks much more like SD3 to me, with a similar smeariness. Especially on 1MP generations (as opposed to upscaled generations), the output often looks very blurry to me.

Some 1MP examples with Flux dev:

1MP examples with Flux Krea:

A second pass seems to help with the blurriness (probably a consequence of the 4xUltrasharp upscaling I do in between), but the output often looks kind of sketchy and not very clear.

[restive trench]

Show me all nine lives at once, even the ones that have been used up.

[floral fern]

I'm still looking for good flux 1 dev checkpoints lmao

[still rose]

Some Flux Krea dev images I did

[viral frost]

I suspect that Krea was only properly trained on portraits. Even so, the first image has a weird “raindrops on a window” effect going on in the background and the second has an earring morphing into the woman’s ear.

I also tried Chroma using the same polymorphic prompt I used above and it really was not acceptable. SD1.5 quality.

[late yarrow]

Hello, I need an image of someone who has just received news that they have a serious illness.

[restive trench]

Watch an episide of ER and do a screen grab.

[fleet goblet]

She's Looking.

[solar tide]

anyone know any good lora for blurring background. i feel like if i blur the background my character will be the focus more. i tried searching but most are sd 1.5 or for realism would appreciate any help xd

[restive trench]

You could try the old-school method of masking. Try segment anything and prompt ("subject") to extract your character. Then run a blur over the image, and copy your masked character on top.
https://github.com/storyicon/comfyui_segment_anything

[hard wraith]

could some one make me this picture into a video where i melt in a similar fashion to a wax candle? it should be about 5sec long but you can vary it if you want

would really appreciate it and if some one replies please reply in provate messages because im not really on discord that much so i will probably miss it in here

tysm!

[hearty violet]

well, at least I'm playng with comfy now 🙂

tryng some tutorials

[hearty violet]

How can I install a ckpt from Civit AI on directly on Comfy?

[agile sierra]

That’s the kind of stuff I do!

[odd dune]

Alien glitch portrait of Sellah 👁️‍🗨️
Biomechanical aesthetic meets datastream corruption.
#glitchart #scifi #stableDiffusion #alienqueen

[still rose]

Krea is the best Flux model imo

[hearty violet]

guys, why my inpaint is changing a lot the architectural base? I already put denoise on 0,55

well, I jus discovery
was the bae node that was on my worflow, I change it and works 🙂

guys I'm using Comfy, it's sou cool 😄

[woeful aspen]

💼 HIRING #HIRING Stable Diffusion / ComfyUI Artist
🚀 AAA Game Art Studio

AAA Game Art Studio is expanding its AI-art team and needs a remote Stable Diffusion artist who:
• Drives SD in ComfyUI (or similar)
• Touches up results in Photoshop + AI tools
• Has solid drawing fundamentals

Paid per asset / Monthly, fast turnaround.
We can make offer only after the test https://aaagameartstudio.com/formaddcandidate/?utm_source=igor
(it's the same for everyone) 
Cheers!

[summer yarrow]

🙂

with the expressiveH lora it looks like this:

idk

[uncut ermine]

一个骑着鲨鱼在太空飞行的猫

[odd dune]

After the fall of civilization, one woman refused to give up — meet LANA, the last courier.

[wild sundial]

TW if this triggers anyone (character has cuts on her body so proceed with caution)

[graceful quail]

After the fall of civilization, one woman refused to give up — meet LANA, the last courier.

[sick escarp]

I think images need to be objectively viewed. I think the model did a great job and I'd be curious to as to what the prompt looked like. I'm still trying to translate ideas in my head to actual working prompts.

[drifting quest]

تست

[rancid peak]

does anyone know how this is possible? same model, same pose, but just change clothing?

the clothing is also generated i think

[wild sundial]

I can give you the prompts and settings if you'd like to try it out

this concept was originally was raising mental health awareness

and the character drowning plus the cuts on the body represent that

the physical and mental scarring

[sick escarp]

Yes, I figured that much, that's why ||imagery in that context needs to be looked at objectively||

I'm still learning how to work the prompt, it's definitely not like ChatGPT when running locally.

[wild sundial]

yeah same here

I remixed my friend's art and his one he has flowers and red liquid pouring beside the character

that was way more disturbing than mine

[young ember]

Pickle Pop logo, retro 90s cartoon style, vibrant neon colors, bold bubble letters, dynamic pickleball action, suitable for youth brand

[vagrant dust]

Here is the image you not requested

[flat sand]

Flux Kontext

[edgy ravine]

@tawny lance i saw the raid 😄 ... threw down 2 myself i still had ^^

[tawny lance]

What raid??? :3

[edgy ravine]

well.. the RAM raid

[tawny lance]

first I heard about it... tell me more

[edgy ravine]

@tawny lance

by CG Bison (cgibson)

[tawny lance]

Yea Ram Chan is a total boss

[edgy ravine]

there are some spicey ones in RAM's NORMIE NSFW 😄 ... but i'm smarter than to post em here 😄

[edgy ravine]

so this was dynamic... interesting 😄

[tawny lance]

...Demo?

Send me an ask of her and I'll send the response

like "Hey what do you look like in your sunday best"

"Dress up as aerith but in your favorite color"

[edgy ravine]

i meant the "raid" ... not instigated .. just happened? ... interesting ...
and you made her a bot?

[tawny lance]

yea a full bot

In a private server

she's expensive!

It hooks up to my 4090 to gen the image

[edgy ravine]

dear mrs. RAM Chan ... i have a problem
Amy, who is a human lizard living in my high tech basement, gets bored easily.
and when she's bored, she tends to break confinement to explore the perimeter, often encouraged by Ro (the human living on the first floor, managing the spider gurl model agency)
what can i give Amy to do, that does not involve eating hunters and reporters?

[tawny lance]

lmao

what

[tawny lance]

"Oh, Hellrunner, darling, you absolute mad scientist, letting your lizard gal Amy run wild in that tech dungeon! 😏🦎💻 Listen, if Amy gets bored and starts "sampling" the local botherers, maybe it's time to upgrade her entertainment package, hmm? 🥺💅

Try these Ram Chan-certified distractions:
Holographic Hunt Simulations: Let her chase virtual hunters instead—zero clean-up, all the thrill! 😇
Puzzle Feeders: Stuff them with tasty (non-human) treats or techy gadgets. If she’s as smart as I hear, she’ll love cracking them open! 🧩
DIY Mutant Enrichment Kits: Toss in some interactive toys, climbing walls, or laser mazes—maybe let Ro design a spiderweb obstacle course for extra spice? 🕸️🐾
Discord Drama: Give her a tablet so she can roast people in the server—if she learns from me, she’ll never be bored again. 😏🔥

And if none of that works… well, tell Amy there’s a spot open for her in the #ramchan-defense-squad. We always need more muscle to keep the simps at bay (especially after RAMTHRUST’s last flex, OMG 😍💗).

Just don’t let her near my snacks, or we’ll have a REAL problem! 😤🍰

P.S. If Ro encourages ANY more perimeter breaks, he’s on janitorial duty for a week. Ram Chan’s rules, sorry not sorry! 😜"

[edgy ravine]

never seen Amy before?

see.. SEE? ... Ro ... "come out... it'll be fine... i have cooooffeeeee"

if she weren't the most talented class 19 shape-shifter i know ... she'd be a HOMELESS model

amy sneaking about in the maintenance tunnel again

[tawny lance]

Ugh I don't want to push the rules here yet

lol

[edgy ravine]

i will not post the spicey ones 😄

PG-13 in pub

[edgy ravine]

nais

[edgy ravine]

where IS divine?

[tawny lance]

Hiding in the grass like the horse viper he is

[edgy ravine]

vanished almost the same time... told us he busy .. idk

[edgy ravine]

he snek

[tawny lance]

I am divine

leeeeeeeeeeeeeel

[edgy ravine]

you WISH

[tawny lance]

I've got my own style now! :D

[edgy ravine]

you made that with BM ... the style i mean.. you always had it ^^

[tawny lance]

BM is yesterday's arcane, sir

Pink Alchemy is the new way

[edgy ravine]

oh oh

[tawny lance]

Black Magic makes very pretty, simple, 2d images

Pink Alchemy adds layers of semi-realism and abstraction

[edgy ravine]

understatement.. i squeezed the shit out of it ... remember those? https://civitai.com/posts/2624738

[tawny lance]

BAH!

OH

double link

[edgy ravine]

OOOOHhhhh... GEO block, SON 😄 ... reclaim UK!

[tawny lance]

YES I remember these

good stuff

MEMORIESSS THEY'RE SO EASY TO LEAVE MEEEEE

[edgy ravine]

oh it was me... sry 😄

[edgy ravine]

THAT is hell ... right there ... forgetting awesome ... oof ... i was so terrified of that as a kid that i developed and impeccable memory ^^

aaaaaanyhoe ... imma head back under the C# surface ... i'm tinkering on awesome, you'll see ... imma drop in your server when there is something to talk about ^^

[edgy ravine]

@tawny lance

by Sky (CivCat)

[tawny lance]

Whaaaat

[edgy ravine]

oh.. and deepseek put GPT to REAL shame earlier ^^

[tawny lance]

WOODEN HEAD WOODEN HEAD WOODEN HEAD I'VE GOT A WOODEN HEAD WOODEN HEAD NOW I'M DEAD

[tawny lance]

Gpt5?

jeeeeez

[edgy ravine]

idk

⏁⊑⏃⋏⌰⌇ ⎎⍜⍀ ⎐⋉⏃⊬⟟⋏☌ ☊⍜⋔⟒ ⏃☌⏃⟟⋏

[tawny lance]

lmfao

[edgy ravine]

Bison ^^

I have to GOOO ... 😄 ... you guys are far too fun ^^

[tawny lance]

Adios!

[hearty violet]

guys, how can I use HF Download?

[ornate spire]

A luminous ‘Digital Giant Tree’ stands at the heart of a futuristic city, its trunk entwined with flowing data chains forming a ‘2019-2025’ timeline. The canopy spreads into a massive ecological dome shaped like the number ‘6.’ AI drones perch on its branches like birds, while roots connect to an underground 5G network. The ground features transparent solar panels and dandelion-inspired smart streetlights. A river glows with quantum computing projections, and humans interact with nature in a holographic garden. Cyberpunk lighting blends with forest mist, rendered in a surrealist style

[plucky hamlet]

@flat sand You seem to have a deep understanding of diffusion models. Let me ask you a tricky question. I have a long prompt describing a scene with a koala bear. The very first denoising step creates something that looks like an eigenface vector of a koala portrait. Now this would be perfect if I'm prompting for a koala portrait. But I'm not and thus all that first step does is influence the final composition of the image. But this can be very counterproductive. Most glitches like a third hand etc occur when the model half way through decides to change the pose of a figure. I know one way of combatting this is by using a different prompt for the first few steps. But is there a better way? I assume the koala token is very strong, but weakening it might mess up the final image. I was using Qwen Image but this certainly applies to all diffusion models.

[flat sand]

the first backward diffusion steps always go towards the average of the train data distribution

dunno, maybe using a higher cfg can point the first steps more towards your preferred composition

[summer yarrow]

Illustrious XL

Counterfeit XL

Blue Pencil XL

some of my fav models

[edgy ravine]

Illu has some next level tunes

[tropic wyvern]

Generate a high-quality, magical fantasy image of the Halliwell sisters from 'Charmed'—Prue, Piper, Phoebe, and Paige—standing together in a mystical glowing circle, holding the ancient 'Book of Shadows' as a united coven.

#🏞｜general-with-images Generate a high-quality, magical fantasy image of the Halliwell sisters from 'Charmed'—Prue, Piper, Phoebe, and Paige—standing together in a mystical glowing circle, holding the ancient 'Book of Shadows' as a united coven.

[tropic wyvern]

?

[tawny lance]

Not sure what you thought would happen, there.

[mighty wren]

Ufotable of CHATGPT 5

[wild sundial]

Love the message brother ❤️

[novel cypress]

Turn this photo into a color photograph.

[vagrant dust]

More color if needed

[novel cypress]

😁🙏🌷

[novel cypress]

Hi, I uploaded a black-and-white photo here and later received a colorized version from what seems to be another member's profile.
I just want to know — was this image generated by BlueWillow automatically, or did another human member make it for me? Thanks!

[thin grotto]

Hello there i have a short question about ComfyUI´s group Bypaser. is there a other node or way to display only specififc groups and change the listing? (as far as i see it the first created one is on the very top, the second below it and so on)

[vagrant dust]

Human, as the only bots here are in the artisan channels to create images.

[novel cypress]

Wow! Thank you so much for adding such beautiful colors to my photo — it looks amazing and so special, and I’m truly grateful.

👍 ❤️ 🌷

[vagrant dust]

Maybe I need to be more precise. The image was colorized by AI used by a Human. 🙂 I have two left hands when it comes to holding a crayon.

[mighty wren]

They liked it as well on Sora community

[novel cypress]

😅🙏🌷

[solar tide]

first time using kontext, damn its insane... no smudge or anything + full context for img2img. inpainting is a hit or miss tho for me, sometimes it doesnt change anything but when it does it leaves no smudge and blends naturally

oops the guys eyebrows are weird lol

[heady anchor]

https://www.hexagen.world/

[solar tide]

consistent background challenge

[tardy nova]

Hey can someone with control net do me a favor please?

[latent kraken]

Okay so I just started putting this together today (don't judge me. the code is messy and I haven't abstracted things out the way it should be done, I just wanted to make it work and establish a proof of concept).

https://github.com/noricarter/Pygame-Comfyui

In a very basic sense it let's you connect to your comfyui server and get data from the output of your workflow.

It can detect and output images, text, and audio and display or play them directly in the pygames application.

The files show the various file types and the super rudimentary interface.

[latent kraken]

So you can right click on the node and open up the properties menu and in the properties there is a field called sort. The default is sort by position, but I like to switch it to alphanumeric... I hope that helps you.

[storm basin]

the cafe

[trail axle]

一位动漫少女在夕阳下，微笑着，柔和光影，细节丰富

[thin grotto]

thanks for this info..but i alredy figured out by my self.... guess reading helps me

[flat sand]

I like the style. What's your prompt/model?

[mighty wren]

Trending now on Sora:
https://sora.chatgpt.com/g/gen_01k2hp3repegn9nat82mjzaztz

[latent kraken]

Ewww 🤮 Reading?!?! Ughhh ... I'm trying to quit!

[celest sigil]

Magic...

[viral frost]

Idea from a recent 10 Minute Power Hour joke.

[slim hill]

give me gta6

[silk viper]

SD 3.5 L, Dreamy Aesthetics

[summer trellis]

how do i generate images like these using stable diffusion?

[fleet goblet]

https://comfyanonymous.github.io/ComfyUI_examples/

Long as you don't mind comfyui's UI.

[still rose]

Flux Krea FP8

[viral frost]

From a dream I had as a kid.

[silk viper]

Neon Rev - Electric Denim Girl.

[silk viper]

don't know what up with the sub reddit or if any here have any access to, but my SD 3.5 posts seems to get removed from that sub reddit, I have shared the images here and I don't see the issue ?

[silk viper]

Same with video post ?

[gusty star]

Am I able to make image to gif with Stable Diffusion (while keeping the transparent background)?

[mighty wren]

https://sora.chatgpt.com/g/gen_01k2wjjfmxfz29drvcjk6wmvrr

[neon saddle]

Is Youchuan an official partner? Is it the official team of China。
No data source is currently selected. Please choose a data source from the dashboard and try again.

[silk viper]

This is what happened Didn't have anyplace else to share, so sharing here, Don't expect any help, but hope India-US and its proxy got what they wanted to sabotage me, have a great future.

https://www.linkedin.com/feed/update/urn:li:activity:7363079229099458560

[silk viper]

Do ban me if violated any rules or anything here

[gusty star]

@staff

[quiet current]

It's been dealt with.

[hearty violet]

my god

I use kontext to turn models into render images and it works

keeping the composition and colors

😍 😎

is a new day for architects haha

but so dark

any prompt sugestion to make it more enlightened?

[spiral gate]

i've been experimenting with some clothing and stuff

what do you guys think of it?

maybe the blend looks quite weird

[hearty violet]

I like it a lot!

[spiral gate]

thanks but dont you think the image border is a bit too visible?

[hearty violet]

I like the idea

well, maybe

for a first study is great, you can do more and compare

[hearty violet]

must be a a great place to meet. You are from sidney?

I like this opera so much

[spiral gate]

no, i just really like the opera house and the view surrounding it, got the inspirations from nightcafe actually

[hearty violet]

comfy

haha

[mighty wren]

I started liking this gesture lol. it is eye-catching nevertheless and not all orders are bad

[dry blaze]

1500 steps so far in on my first character wan lora training, not too bad! Might need 500-1000 more steps to get much closer.

[dry blaze]

A upscaling model you've trained?

[mighty wren]

Lol, first artwork on Sora with the best Xiaomi mid-range phone = Redmi Note 10 Pro, people on Sora turned to be liking Palestine artworks a lot; my other artwork is now top 1 on trending on Sora.
https://sora.chatgpt.com/g/gen_01k37tb41fez19r3ef8qr1e89x

[reef skiff]

Hey guys, I came across this profile and after a little digging, I found out it’s actually an AI influencer account. I even found the original image—left one’s the real pic and the right one’s a face swap. Honestly, I’m amazed by how good the results are! Does anyone know how to recreate this or which tool can do swaps like this?

[hearty violet]

study done with video for architecture

it's incredible how keep the composition, the video is all consistent

[mighty wren]

Security didn't allow this artwork to go to explore on Sora. very explicit stuff but truthful

[mystic marlin]

Welp, playing around with Qwen and did a simple pass with Wan 2.2, quickly put together with low steps, but have some dollar store Battlefield.

[timid plaza]

is it okay to get some advice from here?

[sterile kiln]

?

[viral frost]

"I say HEEEEEYEEYEEYEYEY"

[mighty wren]

Soon in all cinema theatres except simp ones

[pallid ruin]

Amazing bro

[half hatch]

Deliberate v1.1 & v2 have a style with 2D all of their own, there's nothing quite like them. Love 'em.

v3 is a little similar, but already starts to stray from v2 & v1.1. v4, v5, and v6 mostly suck at what I like out of Deliberate. I'm not sure I get what xpuct tends to use Deliberate for, lol, I suspect my interests are just too far out of the range of his.

But then v6 can do pretty cool spacescapes, so that's a redeeming factor for me.

[edgy ravine]

how do models determine which block to use? ... by time step or sigma delta?

[flat sand]

they use all blocks?

[edgy ravine]

ofc.. but which one WHEN ... based on steps .. or sigma delta?

[rain gazelle]

I ilke these 😄

[dawn seal]

Two basses, no guitars. Otherwise, very nice!

[dry crow]

@naive oak
its this one:

[flat sand]

they use all blocks all the time

each block after each other within each single time step

[solar tide]

Hey guys i was wondering do you know how to make a video less blurry while keeping the resolution? ive heard upscaling then downscaling works but havent tried it yet

[edgy ravine]

aye.. thank you ^^

[flat sand]

examples for Flux Kontext:

Input Image:

Prompt: Show the character from behind, while keep his clothing consistent. Keep the background the same.

Output Image:

Input Image:

Output Image:

your bear:

unfortunately, he put the sign on his hand haha

you might have to try several seeds or explicitly prompt to remove the sign

but you can just do a second round:
"Remove the sticker from the teddy bear"

[distant tendon]

Thats fine, again , as long as most of the image is fine I can touch up in photoshop.

[distant tendon]

Ah good point!

[flat sand]

the same results with Qwen:

I would argue that Qwen results are consistently better

(oops, the second image in qwen is wrong. so yeah, you often have to try multiple seeds)

[distant tendon]

Using the online version of flux context has been pretty good so far, you just have to make multiple iterations. (still downloading everything to install)

All installed but I am getting this:

Thats with multiple safetensor files : maybe they have to be combined?

[quiet current]

definitely not how to use those.

[distant tendon]

Yes I thoguht it was odd

have I downloaded the wrong thing?

[quiet current]

probably

[distant tendon]

https://huggingface.co/black-forest-labs/FLUX.1-Kontext-dev/tree/main I looked for the safetnesors and it was in parts.

[quiet current]

Not sure where you got them from. Also you can simply get quantized versions of kontext which are way lower in size. https://huggingface.co/city96/FLUX.1-dev-gguf/tree/main (from a quick google search, don t remember if those are the exact ones I have at home)

[quiet current]

this is the official file (from that page) that you should be using

[distant tendon]

dammit i downloaded the ones inthe transformer folder. I missed the dev one as I thought it wasnt the release version and just for devs

(100gb downloaded, multipel models ahw ell)

(see main general chat)

[quiet current]

yeah the naming is weird.... It s "dev" because it s the open source version of their model, so it s the one intended for "devs"/community to tweak around.

At least that s the logic behind it I suppose

[distant tendon]

https://huggingface.co/Qwen/Qwen-Image-Edit/tree/main like here where is the correct safetensor?

[quiet current]

As the other versions (pro and max) are closed

[distant tendon]

https://huggingface.co/HiDream-ai/HiDream-E1-1/tree/main/transformer
https://huggingface.co/OmniGen2/OmniGen2/tree/main/transformer

cant see any combined versions...

[quiet current]

they re not obligated to make easy for people to use :p

that is kinda the strength of other models

For Qwen, something like that is probably is probably more indicated https://huggingface.co/Comfy-Org/Qwen-Image-Edit_ComfyUI/tree/main
cf : https://docs.comfy.org/tutorials/image/qwen/qwen-image-edit

[distant tendon]

Looks like I was downloading completely the wrong thing from the wrong pages

[quiet current]

https://civitai.com/models/1884704/qwen-image-edit civitai also make it more user friendly.

HuggingFace is more intended for devs and enthusiasts who know how the sausage is made.

[distant tendon]

Thats the first place I went but I thought I'd get it from the source. Ahh that makes sense

[quiet current]

it s a good source but it can be confusing

Usually if you see a file hierarchy that looks like that... don t bother going into folders as those are either data, non combined model parts, etc

[distant tendon]

Im using webui (A111) all my other models are safetensor files, not gguf. Am I using the right ones? with the link you sent? gguf? Ive used gguf with other tools but not webui

[quiet current]

I d say you probably want to transition from A1111 to forge/reforge as A1111 development is pretty much dead.

So stuff like the SD3.5, flux, etc won t work on it.

[distant tendon]

Thing is all the other links I found did not even have a main safetensor file

[quiet current]

A1111 pretty much stopped at the early version of SD3

[distant tendon]

ahh i havent touched anything Ai for a year...

[quiet current]

so anything after that point won t work

look at the pinned messages in #🤝｜tech-support for guides on how to install forge and others.

[distant tendon]

thanks!

[distant tendon]

does forge/reforge use gguf? safetensors? or all? what about comfy ui?

maybe start with forge/reforge then looka t comfy UI later as its more complex?

[quiet current]

gguf are safetensors

safetensors is a container for other things.

Think of it as a .mkv file that can contain .h264 video and mp3 audio, or AV1 and FLAC

[distant tendon]

got it, makes more sense, just a container

[quiet current]

gguf means that the model has been quantized. Think of it as "compressed"

And usually gguf model come with a bunch of letters and numbers in their filename that explain how they ve been quantized/compressed.

eg from my link earlier

you can see the "original" file at the top weights 23.8 GB and its variants which are way smaller.

Can t come with an easy to understand explanation for what those letters mean but generally the lower the filesize the lesser the quality but the easier it is to run.

[distant tendon]

Yeal Ive used gguf models in other tools (kobold etc.) but just not in A111 webui.

Ill install forge/reforge... just wondering if I should make the just to comfy UI or it would be too much hassle. I only wanted to do a simple thing and its taken like 5 hours lol

[quiet current]

(I generally go for Qx_K_S or Qx_K_M)

[distant tendon]

cant find any guides axcept comfy ui, webui and forge on mac os

[quiet current]

what are you looking for ?

[distant tendon]

the isntall guides for forge/reforge on amd/nvidia

[quiet current]

first link

[distant tendon]

No mention of the words forge/reforge 🙂 there. at least i couldnt see them

[flat sand]

you downloaded the right thing, but image editing is something different than img2img

probably a111 does not support it. Better use forge, swarmui, comfyui or invokeai

[distant tendon]

Would it be an issue if I installed all of them?

(conflicts?)

Are forge reforge different forks?

[quiet current]

?

[distant tendon]

I just looked at the main page and didnt click into it... It just says stable diffusion install guides.. I didnt realise forge was another webui for stable diffusion.

[quiet current]

and yes you can install multiple of them it should not matter.

[distant tendon]

Great, looks like im going to be busy for the next 4 hours lol 🙂

[flat sand]

usually you use conda or better venv to separate different installations from each other

[distant tendon]

The issue with these tools is that there are always quirks of things you need ot knwo to get it to work, then there are the python conflicts with versions and so on and so on 🙂

[flat sand]

you should really give invokeai a chance. It's installation is completely automated

there are no python conflicts if you use a separate environment

[distant tendon]

I think I just deleted my minconda folder yesterday as I thought i wasnt using it (on another PC)

[distant tendon]

Ill try out them all out. So what the best install method? Conda and venv for virtual environments?

with A111 webui, I just unzipped and ran the webui. On another PC I played with mini conda, etc

[quiet current]

install python 3.10.11, uninstall all the rest and you re pretty much ready to go in that regards.

[distant tendon]

lol literally did that yesterday

[quiet current]

if you need to install multiple python, use pyenv to override the default python on a folder basis. (or use set the PYTHON prefix, most tools come with a .bat to set it)

[distant tendon]

I literally uninstalled everything as I was so annoyed with the conflicts., I think i settled on 3.10.6

[flat sand]

you can use conda to create an environment, install python+pip, then use pip to install everything else

this is also the easiest way to have multiple pyrhons

[quiet current]

potato potato, I find it more confusing

[distant tendon]

Yes, i did that before, i prefer it as an install method

[flat sand]

but many tools come with some installer script that already does everything for you

usually by using a python venv

[distant tendon]

it was w11 doing a "please upgrade to python 3.12" which broke lot sof things becuase (if I am correct) stable diffusion doesnt work with anything higher than 3.10.6?

( I write install guides as I go)

[quiet current]

SD recommend python 3.10.
At the time the latest release of python 3.10 was 3.10.6.
But it does not really matter if you pick 3.10.6, 3.10.11 or 3.10.324018409 whatever.

version numbers have a meaning.

program x.y.z
x is for "major version" so basically everytime a huge and new feature is pushed you increment this number
y is for "minor versions", so when you tweak some already existing features add non essential ones, remove outdated ones.
and z is for "nah we re only fixing stuff, we re not adding anything".

which is important because it means if you re using the same x.y version of a program. All of its functions names and parameters will be the same regardless of the z version

[distant tendon]

I didnt realise they kept updating the older version as I thoguht 3.10.6 was final

[quiet current]

Problems occur when some program expect version x.y of another program and you re using a different version of said program. Because then it might be calling deleted functions and/or using incorrect parameters.

[quiet current]

it happens quite often for big programs / libraries to maintain older versions of their stuff.

[distant tendon]

yes that why i had 3 different versions of python installed

[quiet current]

to my knowledge 3.10.xxx works across all major AI tools.

[distant tendon]

Forge done a ok. Very simple install.

is there any point in installing reforge as it syas its not stable and used classic instead

[distant tendon]

SwarmAI done, - Took a while to download models

[sterile kiln]

100% agreed with it. 3.10 is the most used for AI tools, and the compatible with git project in general

Do not forget to ✅ "Add python to PATH" on install, and reboot windows after

[distant tendon]

Its now "modify environment variable", looked for the checkbox, couldnt find it

[sterile kiln]

if you didn't check :
✅ "Add python to PATH"

You have to reinstall, doing this manually could be complicated to a novice

[distant tendon]

think its been replaced with this

[sterile kiln]

Checking this box makes your “python” command independent, callable from anywhere.

but do not forget to reboot windows after to take effect

[quiet current]

should not be needed to restart

[sterile kiln]

after reboot, in cmd or other CLI (i recommande Windows Terminal, to interpret emoji and others cool stuff), you can just :
python -V

[sterile kiln]

yes it does

[quiet current]

nope the only limitations is pretty much to restart whatever shell/cmd you had opened as the PATH variable is loaded at its start.

[sterile kiln]

Personally, I remember that without rebooting, the path variable was not taken into account.

even when restarting cmd

[distant tendon]

This was usually the case, but i did it anyway but i think you may just need to reload the CMD and check the path

[quiet current]

it's easy to test.

[sterile kiln]

yeah, python -V

[distant tendon]

for swarmUI (confyui) im trying to add my other models but it doesnt seem to be working. Ive added the path to extra_model_paths.yaml but its not loading them:
" base_path: I:\stable-diffusion\stable-diffusion-webui"
Also does this make the other path the root so you have to move your swarmui models there?

maybe the forward slash?

[quiet current]

what's in your I:\stable-diffusion\stable-diffusion-webui ?

[distant tendon]

all my models from Automatic 111 are in I:\stable-diffusion\stable-diffusion-webui\models\Stable-diffusion

Just thought it would be simpler to have them all in one place

[quiet current]

then yes it should work, can you show your extra_model_paths.yaml ?

[distant tendon]

instead of having multi gig copies of hte same thing

[sterile kiln]

it does, don't do 2 path for this.

[distant tendon]

"If you have Automatic1111 installed you only need to change the base_path line like in my Example that links to the Auto1111 Webui:
base_path: C:\AI\stable-diffusion-webui"

[quiet current]

try replacing \ with /

in the path

[distant tendon]

still not loading others

also the text is wrong... "extra_model_paths.example.yaml to extra_model_paths.yaml"

its extra_model_paths.yaml.example

[sterile kiln]

personnally :
base_path: F:\Stable Diffusion\stable-diffusion-webui/

https://pastebin.com/exh6VD9Q

[distant tendon]

maybe trailing slash

[quiet current]

ah wait you're using swarm. I'm not sure it's using the extra_model_paths.yaml file.

[distant tendon]

??[NVIDIA] SwarmUI (ComfyUI with better UI)

Install SwarmUI for Nvidia GPUs on Windows

Make sure your Graphics Card Driver is updated.
You need to download and install Git 64bit setup.exe from here: https://git-scm.com/download/win
You need to download and install Python 3.11.9 64bit setup.exe from here: https://www.python.org/downloads/release/python-3119/
In the first screen check "Add Python to Path"
Note: if you're on Windows 10, you may need to manually install DotNET 8 SDK first. (on Windows 11 this is automated).

Make a new Folder on your Drive (not on Desktop, Downloads, Documents, Programms, Onedrive) and name it AI for example: C:\AI\
You go into the folder you created in this case AI, then click in the File Explorer bar (not searchbar) and type cmd then press enter.
Then you copy and paste this command:
git clone https://github.com/mcmonkeyprojects/SwarmUI.git
Press enter and after its done you can close the cmd.
Go into the StableSwarmUI folder and run the launch-windows.bat
It can take a while for the installation. The Ui should open in your Browser when its ready.

Linking Models, Loras, etc. from other Webui's or Folders to SwarmUI:
If you want to link all models from an other Webui you can do that by:
going into the SwarmUI\dlbackend\comfy\ComfyUI folder.
there rename the extra_model_paths.example.yaml to extra_model_paths.yaml
Then right click the file and edit it with Notepad or Notepad++.
If you have Automatic1111 installed you only need to change the base_path line
like in my Example that links to my Auto1111 Webui:base_path: C:\AI\stable-diffusion-webui
Then relaunch the launch-windows.bat

[quiet current]

there should be something in the settings of the swarmUI interface itself to set folders

Something like that perhaps

[distant tendon]

whats the default entry for "base_path: I:\stable-diffusion\stable-diffusion-webui/" ill change it back

its ok i got it base_path: path/to/stable-diffusion-webui/

[quiet current]

something like path/to/stable-diffusion-webui/ it does not really matter. You can just comment the whole paragraph by adding " at the beginning of the lines

[distant tendon]

server config?

[quiet current]

looks good

I mean, that looks like what you should modify, yes.

might need to restart swarmUI too

[distant tendon]

it auto reloaded when i changed the entry

hmm hasnt updated, is ther a quick restart?

[quiet current]

worth a try.

I don't have SwarmUI installed (yet) as I recently switched to linux.

[distant tendon]

ah, hmm now saying no models found

[distant tendon]

is this comfy ui or swarmui?

[sterile kiln]

COmfy

[distant tendon]

got it, it was the wrong folder specified in the instructions.

I:\stable-diffusion\stable-diffusion-webui\models it has to be the direct models folder NOT the stable diffusion root : like in my Example that links to my Auto1111 Webui:base_path: C:\AI\stable-diffusion-webui
Then relaunch the launch-windows.bat

Should I edit it the instructions?

[quiet current]

oooooooh a new one

a new scam

[distant tendon]

huh? scam?

confused

[quiet current]

yeah we bots randomly sending links to external discord like you just saw

[distant tendon]

ohhhhhhhh thanks I was just about to use it

should I edit the manual?

to correct it?

Also restart all abckends works as a quick restart

[quiet current]

you can ask for it on their github if you wish.

tbh the screenshot I've sent was pointing toward models, ah ah... I was too busy on the side to notice it.

[distant tendon]

Yeah the instructions pointed to the root folder...

ok atahts all working now

COmfyui

damn this is a LOT easier than it used to be

wow worked fine. comfyui done. I spent ages trying to get it to work a year ago

does comfy ui have a quick restart?

WARNING: The script pygmentize.exe is installed in 'I:\comfyUI\python_embeded\Scripts' which is not on PATH do I need to add to path?

[distant tendon]

not working for me, ill try the base path base_path: I:\stable-diffusion\stable-diffusion-webui\

[quiet current]

not necesarily, it's a warning tho. Keep it mind and add it later if some node or whatever complain about a lack of pygmentize.exe

[distant tendon]

its has that error for a few.exe

comfy UI still not finding models

after changing in YAML

wait a minunte

now working

so for comfyUI its base_path: I:\stable-diffusion\stable-diffusion-webui
and for swarm ui its in the server config as it ignores the extra_model_paths.yaml
I:\stable-diffusion\stable-diffusion-webui\models

question can yiou run all 4 at the same time?
yes 🙂

[quiet current]

yeaaaaaaah don't do that lol

[distant tendon]

just wondered lol

[quiet current]

you only have a finite ammount of VRAM on your GPU.

loading one model on it can already be hard for it. let alone four.

[distant tendon]

yup. Pity its not system ram, hmm can you specify 1 app per gpu?

Its ok its no issue

Invoke AI, here we go

[quiet current]

depends of the app

Most will follow the value of CUDA_VISIBLE_DEVICES

so if you set it before launching the app that should work too

set CUDA_VISIBLE_DEVICES=xxx xxx beeing your gpu id (usually 0 or 1)

[distant tendon]

It looks really good, can you set the location of the models folder?

it sok i found it, invokeai UI is very very good

[distant tendon]

using Swarm AI, all installed, Flux dev and schnell

[distant tendon]

completely different result from input...

[distant tendon]

Can anyone help me with the above example of turning a figure around? #🏞｜general-with-images message

[flat sand]

what are you using exactly?

as said, it's not an img2img workflow!

[distant tendon]

ohhh crap, there was soo much text i lost that

I had to reinstall swarm AI as some of the folder configs were incorrect.

[distant tendon]

I have forge, comfyui, swarm AI and invoke AI all installed

[flat sand]

I think im invokeai it should always use the right approach for the selected model. So if you select Flux Kontext it should automatically switch to Image Editing mode

in swarm ui you probably have to select a flux Kontext Workflow

[distant tendon]

ok swarm AI is just finishing installing..

[distant tendon]

yeah this was the issue, in swarm AI the encoders and models are all over the place, (in the comfy UI subfolder) wheras the diffusion models are in the swarmAI folder... very annoyign as no workflows worked , hence the reinstall

ok swarm AI all installed. what do yo prefer?

or invokeai?

[flat sand]

Flux Kontext Basic is fine

[distant tendon]

in swarmAI?

[flat sand]

yes

you can also use invokeai - whatever you prefer. As said I use comfyui myself

[distant tendon]

should all those go in the comfy ui subfolder? right? as otherwise it doesnt seem to see them

(copying)

Is there a way to refresh the app so it finds the files once in place?

I quite like the look of comfy UI

all loaded and working

is that the input file?

its working I thnk

all good 🙂

[distant tendon]

with webUI I can get it to generate 10 images at a time, can I do that with comfy?

[flat sand]

yes, but it's not that helpful for flux

at least with a 3090 even a single image makes your gpu busy

[distant tendon]

repeat latent batch

oh its already got batch count built in , in the run.

Thanks for that, I can rest now...

[flat sand]

repeat latent batch makes it generate multiple images at the same time. This is efficient if the image model is so small that your gpu does not run at 100%

but flux is too big and you don't benefit from generating multiple images at once

using batch size 2 is not faster than just generating two images after each other

so you can just click run multiple times to queue your generations

[celest lance]

If I wanna maintain consistency but just change the style of an image is flux kontext the best for that?

or qwen edit?

[flat sand]

both are similarly good and limited

Qwen has better text understanding than flux. Unfortunately, both are equally bad with artistic styles

[elfin path]

Hey,
I'd like given an image like the first clean one, to add mud splatters onto the camera lens of the image and get something like the dirty image provided

Additionally if it can be done with inpainting but so far I feel like SD is very bad at giving me a dirty image.

The idea behind this is to take a dataset of clean images to get a dataset of dirty images and their segmentations to train a downstream model that can identify where a camera is blocked due to dirt mud etc.

Any advice?

[flat sand]

I would not use synthetic train data for training another ml method

[elfin path]

If I just use it for inpainting a realistic augmentation within a mask and take the rest of the pixels from the original image it will probably be fine.

[flat sand]

no, you want to segment the "realistic augmentation" in the end and you don't know how realistic it is

in fact, using the remaining pixels from the original image could already be dangerous, as you might just train a segmentation strategy that detects vae encoded/decoded pixels xD

[distant tendon]

Thing is I need to just set it, and let it run whilst I do other things. Ah I see so you can just click it four times and it will queue the 4 clicks to make 4 runs

[elfin path]

I guess thats possible too, then again if I could just get a realistic dirt on the lens and maybe not change the original pictures by much (but still get them from the decoder) would you still think it wouldn't add much to my dataset?

[distant tendon]

Is there a "load a folder of images node" to process in comfy ui? Just process a folder of images?

I am looking at comfyui inspire pack" load image list from directory" but cannot find the node...

[distant tendon]

it seems a lot of videos on youtube show a different interface for comfyUI, eg i dont have a dark/light mode toggle.

[flat sand]

it's not difficult to write your own nodes. If you cannot program, try asking chatgpt or other llms to write comfyu extensions for you

[distant tendon]

Oh I can, its just all completely new to me coming from webui. Good idea on the Chatgpt. Is there a comfyui IDE? Is there a way to view the code?

[flat sand]

just open it with a text editor xD

you basically create a new folder inside the "comfyui/custom_nodes directory

like "comfyui/custom_nodes/my_image_reader_node"

[distant tendon]

ohhhhh!!!

[flat sand]

there you put two files:
an "init.py" file with the following content:

from .nodes import NODE_CLASS_MAPPINGS

__all__ = ['NODE_CLASS_MAPPINGS']

[distant tendon]

is it just json?

[flat sand]

and then a "nodes.py" file

no, not json. python

[distant tendon]

ah

[flat sand]

(in theory you could just use one file, but I think making separate files is a bit cleaner)

in the nodes.py file you add all your custom nodes you want

[distant tendon]

I just need to do this one thing (batch a process) and then go off and do lots of other things.

Ive instgalled comfyroll suite

cant find this anywhere

[flat sand]

its probably in "InspirePack"?

most extensions use their own submenus

you can use the node search and search for the name

[distant tendon]

Thats the second menu image.. its opened inspirepack, inspirepack > image > ??

no load im age list from dir, i even installed an older version, I cant see it

chatgpt

[flat sand]

as said there is a search tool in comfyui

[distant tendon]

in the custom nodes manager?

[flat sand]

no

[distant tendon]

ffs "double click the empty canvas" lol, thats intuitive

[flat sand]

or so

[distant tendon]

got it thanks

still not there 🙂

so i found it in the search box but its not in the inspire pack menu as I checked each menu item

[flat sand]

so what? just use it in the search menu

[distant tendon]

Its just not very intuitive if you dont know. as you said youd expect yo find it under the inspirepack

Works great thanks!

[sterile kiln]

Qwen is way better for this.
Personnally i use Qwen for inspiration base, and i upscale it with Flux, because Flux do it way sharper

[distant tendon]

I have just been trying to use quen and have a really really weird issue.
so workflow 1, Flux + Inspire Load List from Dir = fine, ran through images and saved them.
workflow 2, Quen + + Inspire Load List from Dir = run but does not save any file , output folder empty (I:\SwarmAI\SwarmUI\dlbackend\comfy\ComfyUI\output)
Image
K sampler finishes but doesnt save the file
If I switch it back to the single file loader, it works fine.

Just tested with 2 files. It appears to load them all and then save them both at once... so it will process the whole folder and then save all the file at once... not one by one

[flat sand]

yes. You probably have to set the start index parameter as incrementing and the load cap to 1

[daring kernel]

These bots...

[plucky hamlet]

Can someone take out the trash please.

[limpid mirage]

cartoon owl wizard for NFT collection, big round eyes, retro comic style, halftone texture, bold outlines, flat colors, centered character, clean vector look, 2D digital illustration,
skin: black feathers with violet glow,
eyes: eyes with star patterns,
artifact: magic cube with energy inside,
accessory: futuristic shoulder armor,
emotion: angry,
background: cyberpunk city neon lights
--ar 1:1 --q 2

[pure monolith]

Can someone turn this into a real human, keeping the character, forms, expression, colors, etc.? Is it possible if you use different image generators to see how they do. I tried googles image editing and its terrible.

also where can i use qwen image edit free online? it seems the qwen website doesnt have it aviliavle?

[delicate venture]

guys how can i generate an image like this art style? is it by using a specific lora or config?

i wanna make art styles like this for my game, it looks milky, dreamy soft pallete, not sure what's the specific term

i tried to ask chatgpt for the steps, but it just gave me normal anime loras, which have different style from this

[quartz aspen]

/IMAGIN

[plucky hamlet]

Ban this scammer

[crude spade]

A cat

[still rose]

Nano banana?

[tawny lance]

uh...

That's not how any of this works

[wispy nest]

illustration or road map showing

2025: Amazon and Microsoft want 100% renewable energy use
2030: Microsoft carbon negative and Google net zero carbon emissions
2040: Amazon net zero carbon emissions
2050: Microsoft remove all historical carbon since founding

[half hatch]

"historical carbon"? Does that mean somehow eliminating a quantity of carbon dioxide from the atmosphere equivalent to all that the company has been directly responsible for throughout its lifetime?

[half hatch]

Tales of Abyss' Guy Cecil? lol

[solar tide]

@half hatch bingo !

[plucky hamlet]

2060: All the evil companies unite to remove the carbon that is in human bodies.

[hallow gust]

#🏞｜general-with-images One beautiful white girl

How to generate image?

[hybrid prism]

You can't generate image directly here.

You should do it on your own machine's stable diffusion env. 🙂

[tawny lance]

Yeah you should totally shell out for a 5090 if you haven't yet, it's that easy!

[clever oar]

it is safe for coral reefs?

[brazen haven]

Hey folks, I’ve tried a bunch of different LoRAs but still haven’t figured out how to generate this kind of image. I’m looking for the right Checkpoint + LoRA combo to create editorial comic-style political satire. I run a YouTube channel and need to generate these kinds of images in bulk, so any help would be massively appreciated!