[a / b / c / d / e / f / g / gif / h / hr / k / m / o / p / r / s / t / u / v / vg / vm / vmg / vr / vrpg / vst / w / wg] [i / ic] [r9k / s4s / vip] [cm / hm / lgbt / y] [3 / aco / adv / an / bant / biz / cgl / ck / co / diy / fa / fit / gd / hc / his / int / jp / lit / mlp / mu / n / news / out / po / pol / pw / qst / sci / soc / sp / tg / toy / trv / tv / vp / vt / wsg / wsr / x / xs] [Settings] [Search] [Mobile] [Home]
Board
Settings Mobile Home
/g/ - Technology


Thread archived.
You cannot reply anymore.


[Advertise on 4chan]


Throwback Gens Edition

Discussion of Free and Open Source Text-to-Image/Video Models

Prev: >>107440649

https://rentry.org/ldg-lazy-getting-started-guide

>UI
ComfyUI: https://github.com/comfyanonymous/ComfyUI
SwarmUI: https://github.com/mcmonkeyprojects/SwarmUI
re/Forge/Classic/Neo: https://rentry.org/ldg-lazy-getting-started-guide#reforgeclassicneo
SD.Next: https://github.com/vladmandic/sdnext
Wan2GP: https://github.com/deepbeepmeep/Wan2GP

>Checkpoints, LoRAs, Upscalers, & Workflows
https://civitai.com
https://civitaiarchive.com/
https://openmodeldb.info
https://openart.ai/workflows

>Tuning
https://github.com/spacepxl/demystifying-sd-finetuning
https://github.com/Nerogar/OneTrainer
https://github.com/kohya-ss/musubi-tuner
https://github.com/kohya-ss/sd-scripts
https://github.com/tdrussell/diffusion-pipe
https://github.com/ostris/ai-toolkit

>Z Image Turbo
https://huggingface.co/Tongyi-MAI/Z-Image-Turbo
https://comfyanonymous.github.io/ComfyUI_examples/z_image/

>WanX
https://comfyanonymous.github.io/ComfyUI_examples/wan22/

>NetaYume
https://civitai.com/models/1790792?modelVersionId=2298660
https://nieta-art.feishu.cn/wiki/RY3GwpT59icIQlkWXEfcCqIMnQd

>Chroma
https://huggingface.co/lodestones/Chroma1-Base
https://rentry.org/mvu52t46

>Illustrious
https://rentry.org/comfyui_guide_1girl
https://tagexplorer.github.io/

>Misc
Local Model Meta: https://rentry.org/localmodelsmeta
Share Metadata: https://catbox.moe | https://litterbox.catbox.moe/
GPU Benchmarks: https://chimolog.co/bto-gpu-stable-diffusion-specs/
Img2Prompt: https://huggingface.co/spaces/fancyfeast/joy-caption-beta-one
Txt2Img Plugin: https://github.com/Acly/krita-ai-diffusion
Archive: https://rentry.org/sdg-link
Bakery: https://rentry.org/ldgcollage

>Neighbors
>>>/aco/csdg
>>>/b/degen
>>>/r/realistic+parody
>>>/gif/vdg
>>>/d/ddg
>>>/e/edg
>>>/h/hdg
>>>/trash/slop
>>>/vt/vtai
>>>/u/udg

>Local Text
>>>/g/lmg

>Maintain Thread Quality
https://rentry.org/debo
>>
I'm feeling extra adventurous today. think i might try out a "1girl, sitting" gen
>>
>nigbo
>>
Comfy must be dragged onto the streets and shot
>>
>another debo thread
any anon baking?
>>
Thank you for the nigbobumps.
>>
so when we arent getting the base model today, can you finally admit that it isnt coming?
>>
it's funny how there's always a nigbo and comfy getting shot in the beginning of every thread lol
>>
File: 1764895084289269.png (396 KB, 669x576)
396 KB
396 KB PNG
Only the best for my anons:


https://files.catbox.moe/g1wpjn.7z
>>
>>107444670
I'm not sure what you mean, but if you make a better bake I'll happily delete this one
>>
what a fucking dumpster of a general lol
>>
>>107444687
you should try some self reflection
>>
>>107444702
he keeps trolling because people like you take the bait lol
>>
>>107444687
based, moar bbc is always appreciated
>>
Give me some ideas for gens?
>>
File: 1764897161861219.png (607 KB, 730x1024)
607 KB
607 KB PNG
>>107444702
Rice bunnies love Black Cock™
>>
File: Nano Banana Pro.jpg (809 KB, 2000x2000)
809 KB
809 KB JPG
Wake me up when a local model can do this shit
https://www.reddit.com/r/nanobanana/comments/1p629vg/i_made_a_set_of_examples_using_the_nanobanana_pro/
>>
Hello /bestgeneral/
>>
which side of the spectrum /ldg/ likes?
>>
Back to baking before the image limit I see. Z hype lasted less than a week. Local really is a complete joke
>>
>>107444736
>Local really is a complete joke
always has been
>>
>>107444735
i hate all and every woman
>>
>>107444735
yes
>>
>>107444750
>>107444736
Just because 4chan and this thread attracts retards doesn't mean anything else.
>>
>>107444736
the hype died because of the Z-image base doom and gloom, people are not in the mood when they have a suspicion Tengyi will say at some point they won't release the 2 other models
>>
File: 00459-3556859233.jpg (579 KB, 1536x2688)
579 KB
579 KB JPG
H-h-hello... is, um, i-is local anime h-here...?
>>
>>107444736
Its always the same with every new AI model that get released, people get hyped until they get bored of its limitations
>>
what causes this?
>>
File: Z-image turbo.png (1.27 MB, 1024x1024)
1.27 MB
1.27 MB PNG
>>
File: ComfyUI_07585_.png (1.79 MB, 832x1216)
1.79 MB
1.79 MB PNG
>>107444765
only the best local 1girls, standing
>>
File: 112719984.jpg (251 KB, 832x1216)
251 KB
251 KB JPG
Don't look at me like that, Onīsan... please...
>>
>he's samefagging again
Kek'd.
>>
File: 00031-1946038050.png (2.34 MB, 1248x1824)
2.34 MB
2.34 MB PNG
>>107444736
got bored of z image and wet back to illustrious.
>>
File: 3531881835.png (992 KB, 896x1152)
992 KB
992 KB PNG
o o
P <A> o
/I / \ / |

>>107444735
I like them all.
>>
>>107444798
It's easy to get bored if the only thing on your mind is 1girl portrait with big bobs and vegana.
>>
File: 00466-3655922384.jpg (435 KB, 1536x2688)
435 KB
435 KB JPG
Already bored with ZiT, returning to SDXL, just werks
>>
>>107444812
What's on your mind? 1girl portrait with flat chest? Jew memes?
>>
>>107444813
Fuck you ranfaggot, stay in your discord
You and nigbo deserve each other
>>
>>107444719
its crazy how we're gonna get a local model that can do this in a year
>>
>>107444813
based
>>
>>107444736
>Z hype lasted less than a week.
No one is gonna settle for a distilled model in the long term, we need the base model, what the fuck are those chinks waiting for?
>>
File: file.png (73 KB, 893x732)
73 KB
73 KB PNG
why did everyone start doing this all of a sudden?
>>
File: 00479-4099899080.jpg (369 KB, 1536x2688)
369 KB
369 KB JPG
>>107444812
what else am I supposed to if not genning 1girl?
>>
>>107444839
>new = good
they don't care if it's the least convenient piece of shit ever, their only job on twitter is to get hype so that they get a better salary
>>
>>107444839
"so they can feel like they're coding" my nigga we are literally trying to create ai goon slop it's not that deep
>>
https://files.catbox.moe/6usx3q.png
>>
>>107444839
AI jeets trends, every ai grifter is grinding the nano banana pro dick, they are trying to squeeze as much ai slop content they can
>>
File: file.png (1.36 MB, 2000x1000)
1.36 MB
1.36 MB PNG
>>107444837
>what the fuck are those chinks waiting for?
You won't get anything, imagine trusting bugs
>>
>>107444872
>1girl, standing
powerful
>>
File: 00039-47212221.png (2.3 MB, 1824x1248)
2.3 MB
2.3 MB PNG
>>107444812
rolling the dice with a high seed diversity is what makes sdxl fun to play with. Zit is too boring and editing a few lines of prompts text barely changes the visual output drastically compared to sdxl.
>>
File: Nano Banana Pro.jpg (945 KB, 2752x1536)
945 KB
945 KB JPG
>>107444872
>every ai grifter is grinding the nano banana pro dick
to be fair, it really looks like the dream model, it has insane prompt understanding, perfect text and has the most pop culture knowledge of all models
>>
File: retroplastic.jpg (952 KB, 1664x2432)
952 KB
952 KB JPG
How do I retro anime but without the plastic look? any tags or loras?
>>
File: ComfyUI_00039_.png (1.33 MB, 1200x1024)
1.33 MB
1.33 MB PNG
>>107444874
trvke, you can only trust them as long as they suck your cock
>>
File: flux2_00010_.jpg (198 KB, 3328x1792)
198 KB
198 KB JPG
>>107444781
>>107444765
1girl standing, tried and true, you love to see it.
>>107444839
My uneducated guess is, since we've moved to actual VLMs within the workflow, people assume that structured input will work better than natural language.
We've seen this in the AI roleplay space as well, where people moved to describing characters using YAML or JSON. Turns out all you're doing is wasting tokens.
I suspect this is similar.
But in the case of image generation, I can see it make sense in case you're describing multiple items or characters. You can make sure that the attributes you're describing are sticking to this single subject, in a structured manner.
But I doubt you'd need full blown JSON for that, any semi-structured input (even markdown) would work to get your point across to the 'prompt optimizer' that most SAAS models use.
For local? Eh, probably just snake oil.
But I'm only talking out of my big old butt.
>>
>>107444839
its a new trick of fighting with the nano banana censorship and improving prompt adherence.
>>
>>107444886
only a un-trained eyes skillet user can read that about that gen, an advanced user like me can see there so much going on, your arms are too short to box with God
>>
>>107444891
>dream model
NBP is insane don't get me wrong, but it has bad style control / no artist styles so I can't do fanart with it, and it's SFW only so I can't jack off with it. Which is why I never use NBP.
>>
File: Z-image turbo.png (1.71 MB, 1024x1024)
1.71 MB
1.71 MB PNG
>Toy Story style, Toy Story style,
>[(Hatsune Miku:0.1):a female character with twin tails:0.3] in a 1950s cocktail dress lounges on a floating couch made of candy canes, sipping a martini that has a tiny city skyline growing inside the glass. The room around her is upside down, with furniture stuck to the ceiling. A parade of ants in bowler hats marches across the walls, carrying glowing lanterns
>>
>>107444863
all shitter leftshits do is strawman a fake scenario in their heads in which they are morally superior to the person they hate (everyone that disagrees with them) and then post it publicly to farm points from similar npcs in their bubble
>>
>>107444945
you remove monetization on twitter and this site will feel normal again, they know rage baiting is the best way to get some bucks
>>
File: ComfyUI_00009.jpg (372 KB, 1536x864)
372 KB
372 KB JPG
another day, another 50 new zit anime loras, we never got this kind of support with flux ,chroma ,qwen or even netaSLOP
>>
>>107444925
>but it has bad style control / no artist styles
that's fair, if it had great style control it could be used to mimic artist styles, but I feel this is a really hard thing to do, no model so far managed to nail that shit
>>
>>107444925
>but it has bad style control
Just wait 2 month when the release Nano Banana Pro 2 bro
>>
>>107444966
Illustrious..?
>>
>>107444980
sorry, by style control I meant style transfer, like you could replicate an artist's style with just one image input
>>
>>107444906
That twitter poster does not know anything. You don't seem to either. Perhaps play with local LLMs more. That being said, 4B is retarded it's useful for creating lists and simple associations, that's all.
>>
Why does zit spell certain words ALWAYS wrong? Like with the video game mag covers that one anon does here, it's always "DEFINATIVE". Like is that a Chinaman's typo or just a some technical quirk of distillation?
>>
>>107444995
Please, elaborate. I'd like to know what you mean, what part did you have issues with, specifically?
>>
File: 1759792805794304.png (151 KB, 465x453)
151 KB
151 KB PNG
>>107444911
nobody cares about your api fart stories. go away, api beggars!
>>
>>107444991
Oh yeah. Well, training a style lora only takes 1 hour on a 4090 if you have the dataset. And (controversially) it doesn't need captioning.
>>
>>107445034
>1 hour
>if you have the dataset
yep that's the problem, with style transfer you get the result in 20 seconds, and you can switch to a new other style and it'll also take another 20 seconds, but like I said, no one managed to make something great with it, I hope it's possible but so far they don't know the trick, not even google
>>
File: Z-image turbo.png (1.92 MB, 1024x1024)
1.92 MB
1.92 MB PNG
>>107444935
it's pretty good at reproducing old anime styles out of the box
>>
>asus gx10
worth it? it's like $3000
>>
>>107444735
the 2D ones on the left of course
>>
Omg the 6b edit model got released!!!
https://huggingface.co/meituan-longcat/LongCat-Image-Edit
https://github.com/meituan-longcat/LongCat-Image?tab=readme-ov-file#text-to-image-generation
>>
File: file.png (146 KB, 640x354)
146 KB
146 KB PNG
>>107445094
>With only 6b parameters, LongCat-Image surpasses...
it's not to late to remove that line unc
>>
File: 112419630.jpg (1.09 MB, 2688x3840)
1.09 MB
1.09 MB JPG
>>
File: 1741676296769536.png (141 KB, 1782x619)
141 KB
141 KB PNG
>>107445094
they claim they beat easily Qwen Image Edit, if it's less slopped that would be great as well
>>
>>107445094
>>107445140
>They didn't include Z-image turbo on their mememarks
:)
>>
I hereby declare /ldg/ to be officially the Anime Website Local Diffusion General . The only non anime gens allowed will be news. Everything else that is 3DPD is going to be considered off topic!!
>>
File: 1.jpg (906 KB, 2221x1988)
906 KB
906 KB JPG
>>107444818
>EVERYONE THAT DOESNT AGREE WITH ME IS RANFAGGOT OR NIGBO OR WHOEVER LIVRES RENT FREE IN MY HEAD!!!!
>>
>>107445094
yea, that looks quite interesting. did you set it up already?
>>
File: 3a7.jpg (81 KB, 680x680)
81 KB
81 KB JPG
>>107445168

>>107424657
>>107424657
>>107424657
>>107424657
>>
>>107445175
>did you set it up already?
if it's not on comfy, might as well not exist
>>
File: 1762203729035.png (1.18 MB, 1280x1120)
1.18 MB
1.18 MB PNG
>>107445168
check this 3DPD
>>
>>107445094
>6b text-to-image model
>trained on 1.2b images (likely largely than Z-Image)
>apache 2 license
>most importantly, not distilled
the chinese saved us from the chinese
>>
>>107445140
>nona
>>
>>107445185
Fuck off, that thread is for discussing the micro stuff and technical process of diffusion. BUT HERE is for the macro stuff and both can coexist independently, so please only post anime or news screenshots.
Your general belong to us.
>>
File: 1760576593785704.png (749 KB, 1557x1530)
749 KB
749 KB PNG
>>107445094
https://xcancel.com/Meituan_LongCat/status/1996950202687918586#m
It's a shame they didn't include Z-imge to the list
>>
>>107445094
>>107445140
whos gonna tell them?
>>
>>107445204
Why do I feel the Tengyi fucks will react to that and pretend that they """finally""" finished the model and release this shit tommorow :^)
>>
File: 00053-1902495651.png (2.64 MB, 1152x2016)
2.64 MB
2.64 MB PNG
>>
>>107445204
Z-image will never be surpassed if the others models are more slopped, let's hope it's not the case for that one
>>
>>107445227
it really wouldn't tell you much more when you have these benchmark scores?

IIRC these benchmarks however don't test questionable/lewd subjects or anime characters or training plasticity
>>
>>107445200
The general etiquette will be to only post anime gens and I don't care what you do, dear spammer.
>>
File: 1741551366856963.png (2.68 MB, 1191x1473)
2.68 MB
2.68 MB PNG
>>107445279
>let's hope it's not the case for that one
https://github.com/meituan-longcat/LongCat-Image/blob/main/assets/LongCat_Image_Technical_Report.pdf
what do you think?
>>
File: ccggik87ibbd1.png (1.38 MB, 1666x2400)
1.38 MB
1.38 MB PNG
https://files.catbox.moe/g9dmem.png
>>
>>107445204
>the chinese saved us from the chinese
imagine if it's true lmao, that one has released everything in one go, no more vague teasing shit, no more waiting bullshit, let's just hope that it's good
>>
File: 112645040.jpg (345 KB, 1536x1024)
345 KB
345 KB JPG
>>
>>107445326
doesn't look bad, but we need to gen our own samples to know if/how usable it is compared to other models
>>
File: 200w.gif (68 KB, 200x134)
68 KB
68 KB GIF
>>107445329
its... BEAUTIFUL
>>
>>107445326
>when slop is not enough
Feels like I'm getting an overdose
>>
File: 1743421270054202.png (3.39 MB, 2472x1270)
3.39 MB
3.39 MB PNG
>>107445326
>https://github.com/meituan-longcat/LongCat-Image/blob/main/assets/LongCat_Image_Technical_Report.pdf
why did they feel it was a good idea to go with that example, LongCat is the worst of them all since it's literally squeezing the ratio of the girl horizontally
>>
File: 1762289727471.png (1.11 MB, 1280x1120)
1.11 MB
1.11 MB PNG
Rice bunnies are running amock, they need to be reminded whos their daddy
>>
File: 00054-1902495651.jpg (634 KB, 1280x1856)
634 KB
634 KB JPG
Onii-san, are… these flowers… for me? But, um, I'm… not the popular girl… Why… would you do this…?
>>
>>107445326
it looks slopped, there's a reason they didn't go for a lot of human examples lol
>>
>>107445329
based
>>
>>107445380
i guess it depends if you can tell it not to do that.
>>
>>107445380
Flux bros...
>>
File: 00057-2367814616.png (2.65 MB, 1536x1536)
2.65 MB
2.65 MB PNG
>>
>>107445140
>>107445227
>No Nano Banana Pro
>No Z-image turbo
kek, it's gonna suck right?
>>
File: ComfyUI_00033_.png (1.55 MB, 1200x926)
1.55 MB
1.55 MB PNG
From my "safe for work" collection, soon in CatBox
>>
File: garbage.png (2.04 MB, 3765x1724)
2.04 MB
2.04 MB PNG
>>107445094
A B S O L U T E S L O P
https://huggingface.co/spaces/anycoderapps/LongCat-Image-Edit
>>
https://github.com/Jonseed/ComfyUI-Detail-Daemon

anyone try this yet?
>>
File: I'm back.png (1.29 MB, 1280x720)
1.29 MB
1.29 MB PNG
>>107445491
fuck... back to the waiting zone we go
>>
File: zitcontrolnet.jpg (1.69 MB, 4096x4096)
1.69 MB
1.69 MB JPG
>>107445506
snake oil
>>
>>107445506
this shit slops and oversaturates the images, the better solution is rescaleCFG but when I tried it on Z-image turbo the difference was so small it was useless
>>
>>107445506
the sigma trick was a thing in the past
it doesn't work well with modern models
>>
File: 1755673467085359.png (1.69 MB, 1024x1024)
1.69 MB
1.69 MB PNG
>>107445094
it's absolutely terrible what the hell
>>
>>107445491
>>107445547
it's examples like this that can make you appreciate how much of a miracle Z-image is, a 6b model is supposed to be some shit garbage like LongCat, only Alibaba found the secret sauce
>>
File: 00270-1001634827.jpg (213 KB, 960x1280)
213 KB
213 KB JPG
>>107445094
This is the z lab?
>>
it'd be nice if there was a node that had all the zimage style presets, as a reference or toggle. cause it can do so many styles but it's hard to know them all at once.
>>
can fl2x tell left from right?
>>
File: 1746447370127480.jpg (932 KB, 2048x1352)
932 KB
932 KB JPG
>>107445094
this is so bad lol
>>
File: 150266688.gif (1.25 MB, 480x270)
1.25 MB
1.25 MB GIF
>>107445513
stop waiting. i'm sure base will be api model lol
>>
File: 1735478370869169.png (1017 KB, 1024x1024)
1017 KB
1017 KB PNG
so netflix are about to ruin movies btw
>>
File: 1764948093090550.jpg (209 KB, 960x1280)
209 KB
209 KB JPG
>>107445586
her arm should be visible (red) otherwise it implies she has a long and deformed arm (yellow)

worst anime slop I seen in some time
>>
File: 1762043800604.png (849 KB, 1200x896)
849 KB
849 KB PNG
>>107445628
>weebs dont even know what real anatomy looks like
checks out
>>
File: image.png (2.75 MB, 1344x1728)
2.75 MB
2.75 MB PNG
>>107445621
Poor guys, I feel bad for them, at least they tried their best.
>>
>>107445621
all i can see here is that it maybe didn't understand the "mix of traditional and modern architecture", the rest doesn't seem too bad
>>
File: ComfyUI_00167_.png (915 KB, 1120x1008)
915 KB
915 KB PNG
>>
File: 1751020676453098.png (972 KB, 1024x1024)
972 KB
972 KB PNG
>>107445626
no ghibli:
>>
File: Z-image turbo.png (1.22 MB, 1280x720)
1.22 MB
1.22 MB PNG
>>107445626
>>107445668
I need context
>>
>>107445672
absolute slop
>>
>>107445672
netflix is buying warner bros.
>>
>>107445628
Hello, for art related discussion and micro diffusion topics, check out the Anime Diffusion Thread >>>/g/adt , thanks
>>
File: 1762493537425120.jpg (1.19 MB, 2048x1748)
1.19 MB
1.19 MB JPG
>>107445621
sovl vs sovless
>>
>>107445667
>anime
now we are talking brother
>>
>>107445695
>>107445221
>>107445168
>spams the general with non-gens while complaining about spammers
broken bugman mind
>>
File: 1758131110712551.png (1.21 MB, 1024x1024)
1.21 MB
1.21 MB PNG
>>107445691
A Netflix movie poster for a movie called "Caesar". On the poster is Barack Obama dressed as Julius Caesar in the Roman senate. Include the text "Barack Obama is: CAESAR". Make the image look like a movie poster.
>>
>>107445717
Powerfull...
>>
>>107445621
LongCat looks fine though? Actually followed the cobblestone streets part. Does kind of look like a mix of traditional and modern architecture, the building sides and windows are modern while the roof and eaves are traditional. The model also attempts to put more detail everywhere vs Z-Image being very simple and clean lines everywhere. If it's not super overcooked with low seed diversity, this seems pretty good.
>>
>>107444891
2 hairy poters and 0 home alones, fucking failure
>>
>>107445561
>the secret sauce
It's called filtering slop from the dataset, which hasn't been tried until now for some reason.
>>
>>107444911
LOCAL diffusion...retard
>>
File: 1757027064856979.png (1.31 MB, 1024x1024)
1.31 MB
1.31 MB PNG
A Netflix movie poster for a movie called "Droyd". On the poster is A bald black man who is half cyborg, holding a white back of powder with the label "FENTANYL" wearing a white tanktop in downtown Chicago. Include the text "George Floyd is: DROYD". Make the image look like a movie poster.

lmao, could easily fix this with reactor face swap but still good.
>>
>>107444911
nano banana is censored? I planned to try it. Is it even worth trying then?
>>
File: ComfyUI_00028_.png (2.33 MB, 1776x1320)
2.33 MB
2.33 MB PNG
>>
File: 1741751161254654.png (1.98 MB, 1024x1024)
1.98 MB
1.98 MB PNG
>>107445751
like so
>>
File: image.png (2.72 MB, 1536x1536)
2.72 MB
2.72 MB PNG
>>
>>107445764
were you born retarded or did you just get that way?
>>
>check /ldg/
>still no Z Image Base or ZIB

WHERE ZIB
GIB ZIB
>>
>finally get around to making my comfyui wrapper for llama-cpp-python
>when using a multimodal projector it doesnt cleanup properly and it starts to memory leak all over the place, even if not passing a single image
sasuga, llama-cpp-python devs!
>>
>>107445786
kek
>>
File: image.png (1.25 MB, 1272x960)
1.25 MB
1.25 MB PNG
>>107445798
We don't need it, we have SDXL
>>
Long vid sisters, SVI 2.0 drop

>Supports both 2.1 and now 2.2
>SVI-2.0 is a unified version: it inherits the motion dynamics of SVI-Film while retaining the stability of SVI-Shot. For example, it enables controllable exit–reenter character attributes and supports subtle scene transitions within a shot.

https://huggingface.co/vita-video-gen/svi-model/tree/main

Apparently some new workflows are still coming but going to see how they fair in the old workflows
>>
>>107445470
yawn
>>
>>107444616
hey faggot. please fucking make the changes if you are going to bake
>>107440673
>>107440684
>>107442351
>>107442360
>>
>>107445796
I'm very retarded, I don't know why.
so, about nano banana?
>>
>>107445725
agreed. at the very least it looks worth trying to see how it does. seems pretty nice in terms of prompt adherence. also in >>107445702
>>
>107445849
Hi Ani.
>>
>>107445849
No? Fuck off.
>>
>>107445872
fuck you nigger comfy grifter shithead pissfaggot.
>>
>>107445893
NTA, but thanks for showing your true colors.
>>
>>107445849
sorry, I'll remove the cumfart links next thread if I bake again
>>
>>107445838
no examples? suspicious.
>>
>>107445796
you think google is going to just let you generate whatever your crusty little black heart desires?
>>
File: 1756166914304818.png (1.96 MB, 1024x1024)
1.96 MB
1.96 MB PNG
>>107445787
>>
>>107445829
SDXL is still the best for anime until Z gets an anime finetune, but I want realism.
>>
>>107445811
you were supposed to fork and debloat cumfart first before doing anything else
>>
>>107445725
>>107445861
the details are terrible and it looks so slopped the water looks like plastic, it genuinely looks like an unreal engine game
>>
>>107445838
>https://huggingface.co/vita-video-gen/svi-model/tree/main

Resolution: The released model is trained on 480p data. It can be applied to 720p generation to some extent, but the consistency is not as strong as that of a model trained directly on 720p data (might be released in the future).
>>
>>107445691
>netflix is buying warner bros.
for 72 billion dollars, this is absolutely insane
>>
>>107445905
Seems to be on a different branch https://github.com/vita-epfl/Stable-Video-Infinity/tree/svi_wan22
>>
File: ComfyUI_00030_.png (2.28 MB, 1760x1328)
2.28 MB
2.28 MB PNG
>>
File: z-l401.jpg (286 KB, 1728x1344)
286 KB
286 KB JPG
>>
File: file.png (146 KB, 1637x675)
146 KB
146 KB PNG
>>107445936
I'll just settle with making a wrapper for the actual llama.cpp binary + generic OAI caller instead of going through hoops to install the fucking garbage that is llama-cpp-python.
picrel (works perfectly too, FUCK shitty python devs). also its not a problem with comfy, it's this shitty python library itself being garbage.
I HATE PYTHON
>>
File: sir.jpg (95 KB, 1920x1080)
95 KB
95 KB JPG
>>107445993
sir, this is the actual spiderman costume, please change the boots color
>>
File: 1747549206040325.png (1.03 MB, 1000x1048)
1.03 MB
1.03 MB PNG
qwen edit is so good, didnt even need to use reactor swap. but if you want a 1:1 face you can.
>>
>>107445811
>>107445988
do you guys not know what llama-server is?
>>
File: 1753484919175907.png (2.6 MB, 1216x1472)
2.6 MB
2.6 MB PNG
Hello, it's anons mother here. I broke into his bedroom after not having seen him in two days, and i found him passed out on the floor with a comfyui queue filled with hundreds of images like this one. Does anyone know what is going on?
>>
>>107446019
can you not read? what the fuck do you think
>wrapper for the actual llama.cpp binary + generic OAI caller
will do?
Point is making the actual execution of llama-server being part of imagegen.
No, I don't have the resources to keep a model running ad infinitum (or rather, not for good moe models) so I'd rather they be started up, gen my prompt, close down (freeing resources) and then go on with the actual image gen.
>>
man this guy is obsessed, his images look like shit too, I don't understand this level of mental illness.
>>
>>107446063
He's a poojeet troon that desperately requires attention because he wants to be fucked by niggers.
>>
File: z-l402.jpg (617 KB, 2048x1152)
617 KB
617 KB JPG
>>107446016
This is powerfull
>>107446063
Why racist?
>>
>>107446090
what does his posting classify as? avatar use? spamming?
>>
>>107446096
spamming obv
>>
>>107446051
oh, i guess that works too.
>>
>>107446096
I'd say avatar use.
>>
If you dont post your own gens now thats spamming, see:

>>107445695
>>107445221
>>107445168
>>
>>107445988
no. comfyui in it's current state is shit. get rid of the fucking retarded telemetry and remove the nodes 2.0 garbage
>>
>>107446119
still better than your shitty failed imgui frontend lmao
>>
just go post anime here >>>/g/adt
>>
>>107446125
ok? have you taken your meds anon?
>>
File: image.png (2.26 MB, 1536x1536)
2.26 MB
2.26 MB PNG
>>107445628
>>
>>107446096
definitely avatar use
>>
>>107446127
>page 8
>>
>>107446151
it's page 2 retard
>>
>>107446142
Nasty. Why would anyone look at images of female butts?
>>
File: 1751465768962206.png (1.35 MB, 1000x1048)
1.35 MB
1.35 MB PNG
https://github.com/Gourieff/ComfyUI-ReActor

all you need to fix qwen edit gens if they change the face (usually fine).

wan 2.2, zimage, qwen edit 2509, open source is so good.
>>
>>107446151
you can bump it

>>107446142
>6 fingers on left hand
You were so close
>>
>>107446160
looks like blurred ass
>>
>>107446168
you think i'm going to upscale this piece of shit, it's just a meme/test
>>
>>107446160
plastic shit, smooth skin, QiE is too slopped to be enjoyed with realistic shit, anime is fine though
>>
File: 8764.jpg (9 KB, 214x235)
9 KB
9 KB JPG
>>107446116
>inb4 clean upscaled at x1.00001 in Comfy
Do whatever lol, I'm just having a good time posting some ontopic banter
Workflow: https://files.catbox.moe/x7ndfy.jpg
>>
>>107446119
>get rid of the fucking retarded telemetry
What telemetry? be specific, link me the code that has telemetry
>remove the nodes 2.0 garbage
It was released too early and still has too many problems. But I believe them when they say that the shitty library they were using for nodes 1.0 had too many fundamental technical limitations and that the new version will be better in the long run.
>>
>>107444888
I like you complain about seed variety when your shit is the exact same image fucking everytime
>>
File: 1738164290604251.png (1.1 MB, 1024x1024)
1.1 MB
1.1 MB PNG
the anime girl is drinking a glass of water.

edit plays nice with zimage gens too.
>>
>>107446188
>6 fingers
>>
File: ComfyUI_00175_.png (1.02 MB, 1200x1056)
1.02 MB
1.02 MB PNG
>>107446187
I like his stuff, its better than z-image slop
>>
Does quality prompts matter for i2v 2.2?
>>
>>107446182
>What telemetry? be specific, link me the code that has telemetry
the login code. you can search Google in the repo where it sends packets to Google servers on startup. also I think the new lib for the frontend is shit and they are going to make annoying features with it to grift comfy cloud so it's not much of an improvement
>>
File: z-image_00414_.png (3.81 MB, 2048x1152)
3.81 MB
3.81 MB PNG
>>107446220
it's an easy fix with resmon and firewall
>>
>>107446198
where the fuck do you see 6 fingers?
>>
>>107446242
>uses windows
>complains about telemetry elsewhere
>>
>>107446250
on the other hand that isn't shown here obviously, retard
>>
>>107446263
kek
>>
>>107446206
>shit genner like another shit genner,
And they say destiny isn't real, match made in heaven <3
>>
>>107446220
Loading an SSO page (if it even does that by default) isn't telemetry dumbfuck. It's like saying the ComfyUI manager is telemetry because it does read-only requests to fetch a list of updated custom nodes. Telemetry is sending identifiable information about your system to some server, not just reading some stuff.

Again, link me the specific telemetry code you are claiming exists.
>>
File: 1737177918900854.png (1.09 MB, 1024x1024)
1.09 MB
1.09 MB PNG
"anime was a mistak-"
>>
File: z-image_00415_.png (3.75 MB, 1152x2048)
3.75 MB
3.75 MB PNG
>>
File: 17143513567585.png (113 KB, 909x1787)
113 KB
113 KB PNG
>>107446220
Where?
>>
>>107444736
>Back to baking before the image limit I see.
? That's how ldg has always worked
>>
File: 1736387437461852.png (385 KB, 643x726)
385 KB
385 KB PNG
>>107445094
>>
>>107446308
just reaching out to servers on the internet is enough to provide metadata at the very least.. who/when/what is captured
>>
>>107446308
https://github.com/comfyanonymous/ComfyUI/blob/35fa091340c60612dfb71cb6822dc23b99a5dac2/comfy_api_nodes/util/client.py#L470
>>
>>107444798
Do you remember how long it took you to move from pony to illustrious though kek
>>
File: z-l403.jpg (1.08 MB, 2752x1536)
1.08 MB
1.08 MB JPG
ZiT generates x10 better comps without anime style tags., I want to try 2 pass controlnet workflow, keeping comp in first pass, add anime tags in second pass but I'm a ComfyLET though. Can an anon share a 2 pass controlnet workflow with two separate positive prompts?
>>
>>107446323
>>107446342
are you trolling? this method is called when handling connectivity errors. it checks if you have connection.
>>
>>107444839
This obvious cargo cult from twitteroids is making me unreasonably angry. JSON prompting in particular, it confirms what mongo in mongodb means.
I think it's mostly because some of those people are making money from this, and I'm not.
>>
>>107446308
wtf is
>secrets.GCS_SERVICE_ACCOUNT_JSON
is comfyui glowing?
>>
File: 1654344583419.gif (1.75 MB, 235x235)
1.75 MB
1.75 MB GIF
>tfw hit the motherload of all doggystyle prompts
>>
>getting gooks
>add caucasian to the prompt
>get trannys
>add transgender and transvestite to the negative
>get gooks again
sigh...
>>
>>107446321
Both are bug people. What's the difference?
>>
>>107446380
its a secret and google would have to kill you if you ever found out the truth
>>
>>107446372
by sending packets to googe? wireshark sees data going out to their servers.
>>
File: zimg_0012.png (2.03 MB, 1024x1496)
2.03 MB
2.03 MB PNG
>>107446372
you cannot pull every retard up the mountain with you anon, there are literal schizos on this board
>>
File: Z-image turbo.png (830 KB, 1280x720)
830 KB
830 KB PNG
>>107445691
>netflix is buying warner bros.
>>
>>107446342
If you weren't a nocoder illiterate retard, you would see that code (which literally just calls a GET request to google.com) is used only for API nodes, when you make an API request, and only if there is some kind of failure, solely for the purpose of diagnosing general connections issues. It's never used if you don't use API nodes (which you can permanently disable BTW).
>>
>>107446398
I think the whole thing originates from the fact if you have api nodes enabled cum ui pings some server to test connection. And then shizo started to spread this around.
>>
>>107446387
>slop bounce lora
>slop twerk lora
>https://github.com/princepainter/ComfyUI-PainterI2V

;)
>>
>>107446380
that's part of a github cicd workflow, it only runs on the remote repo for automations
>>
>>107446395
yes? google's uptime is pretty solid.. but you could probably check against pajeet.com or something instead if you didn't really care about accurately checking if you're online
>>
>>107446409
prompt? how did you get the x/twitter GUI
>>
>>107446423
https://files.catbox.moe/vxjji2.txt
>>
>>107446413
it's called on startup
>>
>>107446127
From what I understand /adt/ anons only browse other anime related generals, maybe you should go there and ask them to bump a bit your general
>>
>>107446380
it's used for automated testing of pull requests. it doesn't even run.
>>107446395
it's the same kind of packet as you send when opening google in browser. There is no content in the packet.
>>107446398
I guess the cause is lack of education+4chan browsing
>>
File: 1762676002386312.png (1.12 MB, 1024x1024)
1.12 MB
1.12 MB PNG
>>107446447
thanks, neat how it followed all that in detail, it's a good model.
>>
>>107446465
>There is no content in the packet
I don't trust it. the test happens regardless if you have API nodes enabled or not
>>
File: zimg_0019.png (2.21 MB, 1024x1496)
2.21 MB
2.21 MB PNG
>>107446416
the app checks if it can reach most people's landing page when it starts up to verify there's an internet connection.

it's open source, change that address to any website you like.
>>
>>107446473
yep, it's thanks to Qwen 3 VL, this shit can eat up to 256k tokens
>>
how about make something that isn't a python shitheap or adding value to cumfart for free
>>
File: zimg_0023.png (2.62 MB, 1024x1496)
2.62 MB
2.62 MB PNG
>>107446479
again, it's hilarious how obvious it is that someone is american
>>
>>107446488
>open source, change that address to any website you like.
He doesn't care. He's been posting about it for over a year.
>>
>>107446479
aiohttp library is open source and it only runs when handling an exception
>>
>>107446127
Is that okay or you need more, anonie?
>>107446516
>>107446504
>>107446492
>>
>>107446380
don't worry about it
>>
File: 90's anime style.jpg (278 KB, 810x1191)
278 KB
278 KB JPG
How long would it take on a 5070 Ti to animate this into a 5 second clip, let's say 539x800 resolution?

If someone with the GPU could try it'd be awesome.
>>
>they don't know I added glowware to popular custom nodes
>>
File: ComfyUI_00002_.jpg (3.11 MB, 4096x4096)
3.11 MB
3.11 MB JPG
>>107446349
>>
>>107446548
a few minutes if you use wan 2.2 and the lightx2v high and low loras.
>>
>>107446558
that's the real risk when the hobby is new-model-new-node-repo every week and "update all" is right there
>>
>>107446569
this guy is seriously mentally ill lol
>>
File: 1755453348888607.png (1.44 MB, 1024x1024)
1.44 MB
1.44 MB PNG
grok is pretty good at generating prompts for meme stuff

prompt: make a stable diffusion prompt with the leader of the Ming Dynasty race swapped with a black man, in a satirical style of Netflix race swapping.

result: Hyper-stylized cinematic portrait of Emperor Zhu Yuanzhang, the Hongwu Emperor and founder of the Ming Dynasty, race-swapped as a proud Black African king in the most aggressive Netflix diversity casting style, dark ebony skin, regal afro-textured hair with golden imperial hair pins, intense dignified expression, wearing elaborate bright red and gold Ming dynasty dragon robes with massive shoulder pads and excessive embroidery, five-clawed golden dragons dramatically swirling, heavy jade emperor’s crown with dangling pearls, ornate silk sash, standing on the steps of the Forbidden City at golden hour, dramatic rim lighting, oversaturated colors, epic wide-angle lens, subtle Netflix logo watermark in the corner, satirical, exaggerated majesty, "historically authentic" in the most 2025 corporate diversity checklist way, ultra-detailed, 8k, cinematic masterpiece, by Greg Rutkowski and Ross Tran

used zimage
>>
>>107446580
already had crypto miners and ransomware attacks because of this
>>
>>107446584
ive never seen someone so obsessed with getting blacked desu
>>
File: Z-image turbo.png (1.54 MB, 1280x720)
1.54 MB
1.54 MB PNG
>>107446548
>>107446585
we wuz acquired and sheet
>>
File: 1745290004758674.png (1.54 MB, 1024x1024)
1.54 MB
1.54 MB PNG
Epic cinematic poster for the Netflix original series "WE WUZ KANGS: Rome", dramatic portrait of Julius Caesar race-swapped as a powerful Black Roman emperor in peak 2025 Netflix diversity casting, deep ebony skin, short sculpted afro fade with golden laurel wreath made of oversized blingy leaves, intense thousand-yard stare, chiseled jawline, wearing blood-red Roman toga with massive gold trim and unnecessary shoulder spikes, golden SPQR medallion the size of a dinner plate, dramatic purple cape billowing in wind, standing atop the Roman Forum at sunset, Colosseum in background, huge glowing title text "WE WUZ KANGS" in bold gold graffiti-style font across the top, subtle Netflix logo and "Black History Month Collection" badge in corner, ultra-saturated colors, extreme rim lighting, lens flare everywhere, hyper-detailed, larger-than-life heroic pose, satirical, exaggerated majesty, "historically empowered" corporate revisionism, 8k octane render, cinematic masterpiece, by Alphonse Mucha and Greg Rutkowski

lmao, grok is pretty good for engineering prompts
>>
>>107446584
>>107446591
he's epically trolling us. just keep responding to his posts
>>
>>107446594
did you prompt 90s anime style? that's pretty good, actually looks like the same 80s/90s aesthetic.
>>
>>107446613
yeah I went for this prompt
>A high-fidelity 1990s retro anime style illustration of a female character sitting in profile on the massive, knobby black tire of a futuristic sci-fi vehicle. The character has short, spiky silver-grey hair, tanned skin, and blue eyes. She is wearing a tight, high-cut white and orange futuristic race-queen bodysuit with the text "REO" printed vertically on the torso. Her outfit includes white gloves, white ankle boots, and a high-tech mechanical collar with wires connecting to a headset visor perched on her head. She is looking to the left, smiling gently, with her left hand resting on the tire and her right hand raised, pointing a gloved finger at a complex orange and white mechanical sensor array hovering above her. The background is a clear, vibrant deep blue sky. The art style mimics Masamune Shirow or 90s cyberpunk anime, featuring bold cel-shading, hard shadows, mechanical detailing, and a vintage promotional poster aesthetic.
>>
>>107446548
Grok does it in 20 seconds. When will local catch up?
>>
>>107446604
>satirical, exaggerated majesty, "historically empowered" corporate revisionism, 8k octane render, cinematic masterpiece, by Alphonse Mucha and Greg Rutkowski
what the fuck is this garbage?
>>
File: 1751085536036316.png (1.57 MB, 1024x1024)
1.57 MB
1.57 MB PNG
grok request: make a stable diffusion prompt with Julius Caesar race swapped with George Floyd, in a satirical style of Netflix race swapping, in a series about Rome with the title "WE WUZ KANGS".

Ultra-dramatic Netflix original series poster for "WE WUZ KANGS: Rome", hyper-stylized portrait of Julius Caesar race-swapped as George Floyd in full Roman imperial glory, exact facial likeness of George Floyd with solemn dignified expression, dark skin, short hair, wearing blood-red Roman senator toga with massive gold embroidery and purple trim, enormous golden laurel wreath crown dripping with bling, giant SPQR medallion necklace, dramatic purple cloak billowing, standing heroically on the steps of the Roman Senate at golden hour, Colosseum glowing in background, massive glowing title text "WE WUZ KANGS" in bold gold ancient-Roman-meets-graffiti font across the top, tagline "I can't breathe... the Senate" in smaller text below, subtle Netflix logo and "Black Excellence Collection" badge in corner, extreme cinematic lighting, lens flare, oversaturated colors, godlike rim lighting, hyper-detailed textures, satirical over-the-top empowerment aesthetic, 8k octane render, masterpiece, by Ross Tran and Greg Rutkowski
>>
>>107446621
how many proxies does this guy have? lmao
>>
>>107446629
>>107446604
very powerfull
>>
>>107446607
>trolling by showing how mindbroken he is
jej
>>
stop feeding the troll, just report and move on
>>
File: 1737726855844436.png (1.33 MB, 1024x1024)
1.33 MB
1.33 MB PNG
>>
0 excitement now. zit just isn't the one
>>
>>107446656
yeah it is, it makes natural looking images without the plastic flux filter, and can do even more. also, no baked in censorship.
>>
>>107446656
>zit just isn't the one
speak for yourself, zit is a special model
>>
>>107446651
report for what? im just posting SFW gens, more SFW than your loli hentai (back to /adt/ btw)
>>
>>107446656
for a 6b model it's really amazing, look at others 6b models and how they weren't supposed to look good at all
>>107445621
>>107445702
>>107446321
>>
File: 1744648374545375.png (1.48 MB, 1024x1024)
1.48 MB
1.48 MB PNG
A vibrant Netflix promotional poster in the style of Pixar Animation Studios, cinematic lighting, highly detailed, title in bold white cartoon letters at the top: "Netflix Presents: DEI SLOPPA", large group of exaggeratedly diverse Pixar-style 3D characters crammed together in a colorful living room, every ethnicity, gender, hairstyle, body type, and fashion choice deliberately over-represented, awkward forced smiles, some characters looking confused or mildly horrified, rainbow lighting, soft bokeh background, big red Netflix logo top left, bottom of image has a large sarcastic quote in bold yellow Pixar-style font with black outline: "Because nothing says 'great storytelling' like checking every box except the one marked 'fun'", ultra-detailed, 8k, pixar render, vibrant colors, dramatic composition, satirical tone, masterpiece, trending on artstation

grok made the prompt based on a few ideas. zimage does it again.
>>
>>107446656
Maybe Z model Base will fix hands problem...
>>
File: ComfyUI_00004_.jpg (1.32 MB, 2848x1600)
1.32 MB
1.32 MB JPG
>>107446691
>>107446656
Forgot gen
>>
File: 1745775642102372.png (1.54 MB, 1024x1024)
1.54 MB
1.54 MB PNG
>>107446688
Netflix promotional poster parody in perfect Pixar Animation style, vibrant cinematic lighting, bold title at the top in white cartoon letters with red outline: "We Need More Black People: The Show", crowded colorful scene packed with dozens of exaggeratedly diverse Black Pixar-style 3D characters all doing completely unrelated random activities at once: one dramatically reading a phone, one holding a pizza box, one breakdancing, one watering plants, one wearing a superhero cape, one playing saxophone, one doing taxes, one crying over a broken lamp, all with big expressive eyes and forced awkward smiles, rainbow confetti falling, bright living-room background, big red Netflix logo top left, bottom center large sarcastic quote in bold yellow Pixar font with black outline: "100% representation. 0% plot.", ultra-detailed, 8k, pixar render, vibrant saturated colors, masterpiece, dramatic composition, satirical, trending on artstation
>>
What's the best hentai AI image program that I can use on a 6GB 1660S?
>>
>>107446699
might have been funny years ago but everything is jeet coded dei now
>>
>>107446704
yourimagination.safetensors
>>
>>107446704
neoforge and a noob model
>>
File: Z-image turbo.png (1.71 MB, 1280x720)
1.71 MB
1.71 MB PNG
>>
File: 1752994066198813.png (84 KB, 1186x453)
84 KB
84 KB PNG
I appreciate elon not lobotomizing his ai models.
>>
>>107446744
ok but local image diffusion?
>>
>>107446744
elon is based, what can I say
>>
File: 1737432377843667.png (1.3 MB, 1024x1024)
1.3 MB
1.3 MB PNG
>>107446754
im using that to make zimage stuff!
>>
>>107446728
>neoforge
>noob
This is bleeding edge tech general retard, go rub sticks together for fire somewhere else
>>
When ready

>>107446763
>>107446763
>>107446763
>>107446763
>>
File: 346747475.jpg (118 KB, 1600x1157)
118 KB
118 KB JPG
>elon is based, what can I say
>>
File: 1757242211157770.png (1.27 MB, 1024x1024)
1.27 MB
1.27 MB PNG
>>107446767
ok there we go. 2 tries.
>>
>>107446767
but you also didn't answer my question
>>107446627

also you posted the saas video model output which is saas only not local
>>
>>107446781
he literally allows millions to shit on indians daily, anon.
>>
>>107446488
Fucking retard, I explained from where the so called rumour originates from.
I don't give a fuck, don't give me advice I never asked for.
Sub 80 iq subhumans like you should just shut up altogether.
>>
>>107446626
when will you have a datacenter's compute in your pc?
>>
Is it true that every new model comes with a free Miggu inside?
>>
>>107445988
https://github.com/KLL535/ComfyUI_Simple_Qwen3-VL-gguf
>>
>>107446560
really nice, could you share a catbox please
>>
>>107445791
nice
>>
last one



[Advertise on 4chan]

Delete Post: [File Only] Style:
[Disable Mobile View / Use Desktop Site]

[Enable Mobile View / Use Mobile Site]

All trademarks and copyrights on this page are owned by their respective parties. Images uploaded are the responsibility of the Poster. Comments are owned by the Poster.