What a shit general

Blessed thread of frenship

>>103880427
Don't get your panties all tangled in a knot.

>>103880098
Please tell me you're only generating adults

Anon on naptime

/sdg/ is a better atmosphere and has better overall posts instead of wading through 200 nogen posts waiting for a good gen to appear

>>103880540
then go back

>>103880540
then why are you here?
go to the other thread and kys

Any good Flux tunes other than Pixelwave? Anything in the works?

Illustrious author uploaded a vpred test of 0.1 Illustrious on HF.
https://huggingface.co/AngelBottomless/Illustrious-v0.1-vpred-test
I guess Illustrious isn't dead after all. But fuck if they are just playing around when v2.0 exists on their hard drive.

>>103880590
i wish there were some examples so I could see the difference
>But fuck if they are just playing around when v2.0 exists on their hard drive.
tru

>no collage
Shameful.

Now the bread is truly blessed

>>103880590
everyone here is a promplet on N00b desu
>>103880777
nice

>>103880590
Model card where?????

death march

>>103880098
>the vid
Pretty sexy, thanks I guess.

>>103880427
>>103880458
>>103880546
>>103880548
You silly boys are SO butt hurt right now!

>>103880777
Where is my gen?

>>103880754
Someone did one on >>>/h/ where I see less details but more varied poses and better lighting with no errors. Probably needs more A/B testing.

>>103880478
Pedos deserve the rope

>>103881206
nice. I take.. all of them. that ok?

>>103881206
the roastier the better

>karras scheduler is designed for 30 steps
I get better results with 50 most of the time.

>>103881206
Imagine the smell

HunyuanVideo sirs, do you ignore this error?
>Token indices sequence length is longer than the specified maximum sequence length for this model (90 > 77). Running this sequence through the model will result in indexing errors

>>103881326
Can't, I have no reference points.

>>103881279
I go ~40 with karras nice gen

So is this whatever hunyan thing everyone talks about usable on 8Gbs like Flux? Is it comfy-only?

>>103881206
Stop bullying ani

>>103881363
Maybe at a very low quant but the results wouldn't be good

>>103881279
30 steps regardless of sampler? But they're widely different, SDE is like 1 step worth 2.

>>103881529
Not even offloading is supported? Flux works with 8bit, just a bit slower than 4.

>>103881279
not using deis/beta in 2025 is blasphemy

>>103881585
this but instead of deis its AYS

>>103881598
Actual comparisons never as I understand, just speculations for meme schedulers.

>>103881585
how many steps/cfg?

genning in 2025 will be all artist knowledge. I haven't even genned in a year, just compiled a list of artists and tagged their styles

>>103881738
tfw anon tries to b8
>>103881746
>genning in 2025 will be all artist knowledge
always has

>>103881746
Genning in 2025 will be owning a 5090

>>103881746
I had a text file with a bunch of artists and styles that I lost a few months ago. I'm still upset

>>103881746
Don't make it look like we need the inkcels, or they'll get even more uppity.

>>103881774
Become an inkcel yourself and you will never yearn for them again.

If you can't run HunyuanVideo you shouldn't be posting here. There's another thread tailor made for your SDXL cope slop

its the year 20seXseX
generating gifs on your 3060 is enough

I like hunyuanvideo gens because they remind me of a few years ago when no one knew what they were doing, it took minutes to get a result, and everything was slop

VRAMlet masterrace

/ldg/ should be for the cutting edge. Vramlets only post their 1girl, standing, bikini, peace signs here because they get no attention in the other thread

>>103881853
Kino

my girls are more cutting edge than your videos surprisingly

>>103881181
I hated it, sorry.

>>103881902
You don't have to be this mindbroken over some gens

>1girl, standing, bikini
>but in.... LE DIFFERENT ARTSTYLE!
This is what sdxl copers consider cutting edge

>>103880777
blessed trips of frenship

>>103881992
At least my SDXL can produce a handstand, unlike your fluxlop.

freckles = SEXO

fennix for you

https://civitai.com/models/12523/pc-98-galgame-elf-retro-semi-pixel-art-style

blood for you

>>103882036
>Flux
Holy shit get with the times grandpa, were making videos now and you're still impressed with 1girl, handstand? Lmao, time to upgrade that 8gb shitbox

>>103882126
Hunyuan can't do handstand tho

>>103882126
>videos
>blurry static shit for 10x the price
Thanks I prefer ultrahd pics for my time=money. You're the one from Coub era with 6 second clips.

>>103881991
I worry for anon he's been at this all week.

>>103882171
Your "ultrahd" slop pictures were boring 2 years ago, it's time to face it you have been left behind and you're coping, just like when Flux came out. I pity you

I hope all anons pooped well today :)

>anon cant post a hunyuan handstand
on behalf of anon i humbly accept anons concession. id like to thank my parents, god, and anon for this esteemed award.

>>103882202
very nice

>>103882210 (You)

>>103882294
anti-sameface that shit yo

>>103881715
for deis at least 40 steps in general

>>103882315
No rebuttal, a desperate plea for me to leave him alone. It's sad really

The list of things noobshit can't do grows. Now it's the date fruits, not just Hasshaku-sama. But the shill won't respond now that the pictures are given as a proof immediately.

>>103882376
>date fruits, not just Hasshaku-sama
i dont care about either of these things tho

>>103882336
I'll see if I can

>>103882376
>Now it's the date fruits
NAI bros it's over

He dreamed of the prospect of genning videos, the future seemed so bright. He was absolutely heart broken when he came to realization it wouldn't run on his 2070 and with each subsequent model release his dream seems farther and farther away. Now he copes with his outdated static anime 1girl slop
There is no creature more pitiable than the vramlet

>>103882392
ok merci pour ta contribution
>>103882474
Did LLM write that for you?

>>103882244
thanks. moar blood!

>>103882457
nice, you know how to automate it? can do it both in comfy and with adetailer with the right commands. worst case, 4 manual inpaints

>>103882495
I have mastery over the English language unlike the vramlet (synonymous with third worlder)

>>103882473
I wouldn't be posting petty shit like that if other models couldn't do it either. But it's only the holy cow of these threads that fails at mundane stuff.

if you dont have 24GB VRAM minimum you are literally a Subhuman in my eyes.

>>103882529
Maybe take control over your burning state first, firstworlder, must've been all the heat from your insane builds.

The artist does not blame the limits of his tools.

>>103882507
I'm using A1111, I can probably use wildcards but I don't like some of the celebrity faces

>>103882578
this
nb4 we're not artists

There's only one hunyuan poster ITT and I'm sure he's not the one who's been seething all last week

>>103882587
life is art

None of you are artists, sorry

>>103882578
>my model can do anything, just post catbox
>k here's pic
>???

No amount of imaginary prompt-fu can put into model what was never there. Hence the coping "ah but I don't need that anyway".

>>103882615
So is troll posting.

>>103882551
This

>>103882606
me? I barely post coz nobody wants to see my peccadillos

I prefer synthologist btw

>>103882622
I am the artist and that whores body is the canvas knowwhatimsayin

>>103882653
I mean granny porn is definitely not appropriate for a blue board

>>103882675
didn't mention porn, is there something you want to share?

And now for something different.

>>103882580
there is a way, here:
and then you'd go like
cute girl, green eyes [SEP]
feisty bitch, brown eyes, (ana de armas:0.4)[SEP]
and so on in adetailer prompt
>>103882695
INVITING

>>103882653
good to know so anon should just ignore those troll posts cause the poster himself can't even gen

>>103882735
Oooh okay I'll try that

Here's another pretty elf

>wildcards
Tell me about these. What are these and how do I use them?

>>103882759
Open the bobs?

>>103882803
You know I can't post NSFW here mate

>>103882803
Speaking of bobs, why does Flux produce the most abhorrent set of tits known to mankind? I mean, they look fine when covered, but if exposed they're an atrocity.

>>103882780
there are a few custom node sets which can do that, I use the impact pack one. you make (or grab) a txt file and when you activate it in your prompt, one random line from that file gets pulled. can be just one word or something more complex. does that make sense?

>>103882845
fluxtastic lora, problem solved. just look at the images ppl made with it, good enough for ya?

I'll just have to use fluxlop for my puns then.

>>103882897
Those are beans, nigga

>>103882831
Sick

>>103882853
>one random line from that file gets pulled
Txt file which you separated in sections (name, necklace, colors)?
Also, where do I link the string node? It doesn't look like a conditioning node

>>103882376
I refuse to believe you're not trolling. Every model has blindspots anon.

>>103882930
I can believe an anime model missing something so obscure as a date fruit (but why does Animagine knows them then?), but to not know Hasshaku-sama, the popular oneshota fetish? Higashi no eden, the best stylish anime of its season, which even the base XL knew to some point, how'd you make an anime model to forget anime?

>>103882920
you make one file for each 'thing', and then you can pick them in the menu. can make subfolders too. that node just outputs the text, there is another one that does the conditioning too

>>103882959
>the popular oneshota fetish
this is news to me tell me more

>>103882959
>Hasshaku-sama
Theres only ~300 tagged images on booru which isn't bad but also not a whole lot to work with

>>103882981
I understand now.

>there is another one that does the conditioning too
Which node is that? One that links the wildcard processor into the prompt?

>>103882895
>fluxtastic
horrible breasts and nipples, SDXL beats the shit out that

>>103882895
>fluxtastic lora
Doesn't do loli, let alone oneeloli. Gens midgets or anorexics

>>103883010
I'm not gonna, but I do wonder if a video exists like this on the internet. I'd like to know (not necessarily see) the closest approximation

>>103883041
'impact wildcard encode', but i prefer the other one and use something else to condition
>>103883068
>>103883079
I don't gen flux (anymore), I just use flux (with that lora if tits are involved) to upscale realism stuff and it beats the living daylights out of anything sdxl, fact.

>>103883093
shutterstock has everything, yes even that

>>103883226
i legitimately have a hard time discerning what's shutterstock and what hunyeanvideo sometimes.
speaking of, sampler/scheduler/steps?

>>103883342
Jesse Pinkman sure let himself go.

>>103883342
Why would he keep posting images of his face?

>>103883342
cutie :3

okay now you can poast the gud gens tho

>>103883342
Look at that lightbulb head shaped lookin' ass nigga sittin' in his gayming chair

>>103883443
sometimes you gotta help yourself, lead by example, and so on. I don't wanna smash the illusion but nobody is posting for you, nobody even knows who you are

>>103883459
>nobody even knows who you are
Of course they do. He is anon.

>>103883423
>Hunyuan knows dates
I kneel

>>103883484
still cant into handstand desu

>models can't into handstand
However, I know for a fact that Illustrious can do upside down pole dancing.

>>103883496
really? fuck.
i'm going back to cogvideo

>>103883496
Nice cope

>>103883516
>cogvideo
almost forgot that existed
>>103883528
not bad at all

My favorite trick in booru prompting remains choosing a slightly dark picture with strong coloring for img2img (100%), prompting a different contrasting color for the clothing, and prompting a third color for the hair or the eyes. You usually end up with a rich mix of hues in the resulting image.

Someone should compile a rentry full of tips and tricks like that, without any basics or how-to stuff. Just tips and ideas for posters who already know what they're doing.

>>103883598
I don't think janny will be

You see, I don't have a problem with the model. I have a problem (if you could even call it that) with the anons using the model. When it first released, there were many interesting and unique outputs; the likes of which no anon had previously seen. Now, it's all meme shit or obviously low effort slop. I want to see UNIQUE outputs. Something that makes me go "wow I haven't seen that before). Not "anime girls talking" or "woman eats thing that looks like penis" or the worst "woman standing but its cool because I trained a lora for her". This new frontier is cool and all but I feel many are succumbing to the same pitfalls that image fags did.
>1girl standing image slop
Yes, I know.

Pissfag checking in. I've done some experiments recently that have fully convinced me that Flux has far more potential than any SDXL tune ever will.

I wanted to generate an image of a girl sitting in the back seat of a car, wetting herself, with an inset of a dam gushing out water for some visual metaphor. Here's a non-cherrypicked example of noobai vpred + a custom lora that helps emphasize all the pee-related concepts: https://files.catbox.moe/b94lg8.png

Meanwhile, here's Flux dev de-destill with a lora trained on the exact same dataset. Flat color anime style lora is mixed in with low weight to try to help the style a bit. Again, first gen and no cherrypicking: https://files.catbox.moe/kcdbna.png

Noobai can't do the dam inset at all. "Inset" is a booru tag, but it doesn't matter, it's usually used for faces and characters. And the text encoder is shit so you can't say "An inset in the top left corner shows a dam releasing water" like I did in the flux prompt. Noobai is also far worse at the consistency of the car interior, and slightly less consistent at the character pose (although you could say it's just trying to be more creative and varied). On the plus side, noobai is way better at style flexibility, NSFW, and broad booru tag knowledge. But of course it is, flux hasn't been trained on that stuff.

Oh, and the flux lora is trained only on tags, no natural language captions, and also trained entirely at 512 resolution. Just a basic lora, on a censored base model. Despite all this, in many areas of prompt understanding it's completely beating the best I can do with noobai, just with a very limited style. If someone full fine tuned flux on 1 million+ booru images, at higher res, on a mix of both tags and captions, it would mog noobai so hard it wouldn't even be funny. I know illustrious / noobai is all the rage right now, but we're really limiting ourselves by not taking advantage of better base models.

>>103883598
bunni

>look up Hunyuan Video
>needs a bunch of loras to get anything done
>none of the loras are tribadism
Pointless.

>>103883788
tonnes of vram too, I would think that's the more prominent filter

>>103883658
>If someone full fine tuned flux on 1 million+ booru images, at higher res, on a mix of both tags and captions, it would mog noobai so hard it wouldn't even be funny.
Has to be on dedistilled because base Flux is impossible to train at that scale, look at pixelwave. It's possible but highly unlikely at this point unless you know some bored anon with a fuck ton of computes who'd do it for the love of the game. A 16ch anime model would in fact be kino you are correct.

>>103883598
What checkpoint/ artists?

>>103883979
https://civitai.com/models/920709/cat-tower-noobai-xl-checkpoint

Sampler was Heun, schedule type SGM Uniform.

I feel uneasy

>>103883850
>some bored anon with a fuck ton of computes who'd do it for the love of the game
Doesn't seem as daunting or expensive as some people would think when I do back-of-the-envelope math. With kohya sd-scripts, on a flux lora at 512 res, I get through ~5000 images total in 3 hours with a 4090. So 1666 images / hr. H100 is more than double the compute, and rents for about $2 / hr. Full finetune is less compute than lora, but if you mixed in some 1024 res images that would be more compute. Let's say it evens out and the H100 is double the throughput, 3333 images / hr. Now, with a dataset of 1 million, that's 300 H100 hours for 1 epoch. $600 per epoch. Illustrious v0.1 was trained on all of danbooru, so 7 million images, for 20 epochs. But flux being a much more powerful model can learn more from less steps. 10 epochs of 1 million images is $6000 with rented H100s by my calculations, and I bet that's enough to start consistently beating illustrious / noobai in many areas.

There's gotta be at least one rich degen (hell, maybe that's me at this point...) willing to drop a few grand to do this.

>>103884098
some thing is a foot

>14th january
>still no Hunyuan i2v
it's over...

>>103884310
for a base model that's fucking amazing, this model is insane on anatomy, who would've known that training a model on porn (that has a lot of complex and various human poses) would help on that huh????

>>103882968
cute

>>103884291
31/01 23:59:59

>>103884479
I'll take it desu, still a promise kept kek

>>103884372
Kek

>>103884479
so... 2 weeks? roughly

>>103884531
>>103884491
I can see it delayed to february, or never released..

iv2 waiting room

>>103884537
at least they should say something if it gets delayed, that silence is defeaning

Give me the MLLM or give me death

>>103884531
>2 weeks
https://youtu.be/9v-33jcEDk4?t=19

i2v has been delayed after the devs saw what the hunyuanvideo lora page looks like on civit.

>>103884560
I tried to warm them but they would not listen

>>103884560
probably adding some extra nsfw capabilities to capitalise

>>103884558
if I had to choose between MLLM and the i2v model, I'd go for the i2v, it's much more important

>>103884579
I think the opposite will happen, maybe they think they uncucked the t2v too much and make the i2v completly cucked and unusable...

>>103884585
Who the fuck fixes up a 'dog like that anyway

>>103884612
roastoids

>>103884602
I don't think there's enough information to conclude anything yet, but by all means speculate

>>103884583
No it isn't. We don't even know the full power of hyvid because we're using a hacky replacement for the text encoder

>>103884617
My father works for the Chinese, trust me.

>>103884620
>We don't even know the full power of hyvid
we know, use their API to see the "full power", give me the i2v first then we can talk about the MLLM

>>103884602
why can't it be a single model that does both t2v and i2v?

When image to video releases will you make funny little video edits of my lowly image gens, anon? Please? Just say you will.

https://github.com/Tencent/HunyuanVideo
at least they're not dead, they checked another box yesterday

>>103884637
because AGI is a meme

>>103884579
imagine the possibilities

>>103884666
They need to ban all celebrity loras on civitai

>>103884154
>(hell, maybe that's me at this point...)
i hope so senpai

Trying to use iopaint but i'm encountering an error
> iopaint.runtime:check_device:67 - CUDA is not available, use cpu instead
Torch is installed
What can be the issue?

>>103884725
You might have to manually install torch, the version with "+cu" or w/e

>>103884666
kek that's really cool

January 2025 will be a sad month:
- The death of tiktok
- Tencent will not release hunyuan i2v

The east has fallen...

>>103884879
We'll probably get a new illustrious tune though

>>103884879
>The death of tiktok
Good

>>103884912
think about the lost coom material

Any ideas or recommendations on models/loras I can use to achieve old school d&d style art similar to this image?

>>103884936
Coom can be found elsewhere, Tik Tok deserved a US ban

Noob question.

If stable diffusion, etc. got taken offline, could I still gen if I had all the stuff and checkpoints downloaded?

>>103885041
No

>>103885041
How did you even find your way to /g/

>>103885041
Sweetie this is LOCAL diffusion general not stable diffusion general.

>>103885041
>If stable diffusion, etc. got taken offline
I don't see how that's even possible, it exists in a million places in a million different flavours

>>103884948
prompt "black and white" and/or search civitai website

>>103885067
>>103885071
>>103885078
>>103885080
Look, I'm completely new to this, and I like my new toy, and I don't want anyone to take it away.

>>103885151
they will never deprive the localchad of his toys we are safe

>>103885151
If "stable diffusion, etc. got taken offline" whatever the fuck that means happened, how exactly do you think that'll affect the models stored on your PC

>>103885151
Ignore the haters, noob questions are welcome. SD is completely local.

>>103885200
don't be a dick, he can read the general title, he knows it's local

>>103885209
>it's local
We all use an INTERNET browser to gen. It's not local

>>103885217
this

>>103885209
maybe he thought it meant we all live in his neighborhood. you can't make any assumptions here, anon.

>>103885217
Checkout this sweet gen I made: http://127.0.0.1:8888/

>hand refiners are still shit 2 years later
>the best way of fixing images is still manual inpaint + krita
come on...

I will continue asking noob questions if I think of them.

>>103885239
Blue board anon

>>103885237
I inferred that he's a dunce, you take it as read
who's being the bigger 'hater'

>>103885239
bro warning next time I don't wanna see your futa diaper porn

>>103885244
The best way is to use a model with 12B parameters but yeah

>>103885244
unironic skill and model issue

>>103885244
>using a hand refiner in the year 2025 of our lord
you're a vramlet who can't use Flux don't you?

>>103885239
This site does not work? Wait if someone knows my IP address can they see my gens?

>>103885267
Flux outputs are boring and rigid as fuck

youd think videofags would be more accepting of vramlets considering they were once that...

>>103885270
not if you have a firewall, most routers/models do
you could open the port and allow the entire internet access, but don't

>>103885278
you can make them more interesting by using loras or an IPAdapter

>>103885248
>>103885256
y'all don't understand art

>>103885281
I'm not accepting of the vramlets, they're the ones who spam this thread with SDXL slop, they got the respect they deserved, none.

>>103885288
This shit BARELY deslopps Flux. Turning the negative guidance up deslopps outputs more even.

>>103885300
I will start mass posting sd1.5 gens unedited and non upscaled because of this comment.

>>103885301
>This shit BARELY deslopps Flux.
not at all, I can get some sovlful pictures with that method, you just need to find the right image input

>>103885313
make sure you use a pass too

>>103885313
>he wants to be respected even less
that's up to you anon, don't come back to cry when people will insult the vramlets again, you fuck around, you find out

>>103885315
You're not helping yourself with that input image but holy shit no way you think the image on the right has any soul. You're insane and probably legally blind.

>>103885334
Show me an image that has soul, I'm sure you won't share it though, you're too insecure to be judged on your taste, I have way more courage than you anon, remember that.

>>103885315
Is the one on the right is flux? Then i was right by dropping it like hot shit
Maybe it will get good in another year

>>103885324
>that's up to you anon, don't come back to cry when people will insult the vramlets again, you fuck around, you find out
you're the one crying :D

>>103885209
How was I being a dick

>>103885342
Now where is the fun in that? My gens are so recognizable that the three anons ITT would lambast me for seething so hard instead of ignoring.
But trust. You've gazed upon my images with lust and envy due to their soulfulness.

How does one prompt a woman pierced through by a sword, Aerith-style? Ideally from the front, but I'll take what I can get. I think it's a very romantic image but near impossible to get with prompting, unless I am missing something important. I have sword, weapon, impaled, death. What else do I need? I'm using Illustrious.

>>103885362
there you go, I knew you wouldn't share anything because of your insecurity, I'm tired of being right all the time you know...

>>103881585
I see little difference with deis/beta vs dpmmp_2m/karras. Is this another meme sampler+scheduler combo?

>>103885367
You should probably speak to a professional instead. One who deals with issues of the mind

>>103881585
beta gives better prompt understanding but it's burning the image, I prefer sgm_uniform, it has the best balance imo

>>103885376
That's real rich coming from the guy genning old ladies fellating hotdogs

>>103885387
kek, he has a point

>>103885376
You're right, it's so fucked up to associate death with romance. I would never.

>>103885246
based newfren

Some simply have better taste. Those who don't will seethe eternally.

>>103885463
So you're seething eternally? Can't be good for your health anon :(

>>103885041
Bruh just unplug your router for a minute and see if you can still gen some shit

How about stop seething

hello,
here is a tarot card
thank you,

Anonymous

Fuck China and fuck TikTok, fucking zoomers kys

does anyone know if the preview options affect VRAM usage at all, or is this something that my ram and cpu handles?

>>103885766
What does it mean though?

>>103885836
你为什么不高兴?

>>103885909
And a ching chong to you as well young lady

>>103885836
>Fuck China
Hunyuan dominates the local video AI space and Deepseek dominates the local llm space, you better lower your fucking tone you fucking westoid scum

>>103885836
it's not China's fault that the west is so woke and cucked all their aps are boring garbage, the only way for the US to keep their relevancy is to mess with China instead of reaching their level of soul, that's petty and it shows how jealous they are

>>103885863
One way to test this is by observing Task Manager with and without the option enabled.

>>103885960
Shut up faggot you'd better being bowing down and praying to you immoral AI overlords, pray your chang overlord lets you get a cent off what your government makes, until then continue selling sewer slop on your street venue

>>103885988
>you'd better being bowing down and praying to you immoral AI overlords
like who? SAI?

>>103886010
>go to China
>they sell me sewer garbage disguised as food
>now I have nigger aids
oh no how could this have happened to me

>>103886030
>sewer garbage
hunyuan and deepseek aren't sewer garbage, SD3.5M is though

Just a reminded how amazing local img2video will be, it's not just wow you can make image move, it also means we can continue any single video from using the last frame of the video.

This means we can actualyl create stories with different prompts to continue for as long as we want, that is a million times more power than random clips with no context or connection to each other like we have now.

Loras will probably work with this too, to finetune futher your needs

Even if China didn't scam 24/7 and sell sewer covid bats to their own people it would still be a hell on earth, no values, no anything just bug people, that's why we have cunny in all their models

>>103886053
Reminder*

>>103886053
anon ai movie waiting room

>>103886053
>it also means we can continue any single video from using the last frame of the video
it doesn't work like that, the last frame isn't enough for the model to get the full context, the transition won't be smooth at all, if you have a video of a guy running, and then you make i2v with the last frame the model won't have any idea what speed should be next

>>103886056
>that's why we have cunny in all their models
Which is based because it lets me prompt all the oneeloli I want

>>103886053
So this means we can not only make a doggy style sex scene, but we can also make Shrek break down the door and see his wife getting fucked by some dude to see his reaction?

Then we can continue that and show Shrek sit down and watch while crying?

Then after that he takes out a gun and shoot 's the wife and the other guys dick?

Then after that he goes to trial and is found innocent? Then the lawyer on that trial has Kardashian like daughters and then we can make a reality show about them?

>>103886056
>no values
you're talking about the west, in the west you can be a transgender girl bat spirit or whatever and people will celebrate that instead of putting that guy on an asylum, that's how you know you live in "no values" land

>>103886074
Thta's not that big of a deal, you can just use a better frame, obviously every single last frame won't be perfect.

Just like movies we can time the cuts to our preference in a way.

>>103886094
you know how steam changed their splash screen of the black dude with the pipe in his head, and when he started to turn it freaked people out
you could do that with anything

>>103886053
>it also means we can continue any single video from using the last frame of the video.
No, it doesn't.

today, i am going to use the superior prompt understanding and vae of flux dev to gen 1girl, standing, photorealistic

>>103886114
based, if i may say, anon

>>103886113
yes it does, that's how img to video works lol

>>103886120
Good luck with that.

>>103886128
Care to explain what you are on about?

>>103886106
>>103886120
it's gonna be worse and worse anon, the more you extend the video through i2v, the more mistake it'll make and at the end it'll look like a fucking mess

>>103886130
two squandered opportunities suggests no
incoming three

can a nigga getta 1girl ova ere

>>103886132
Not if you choose the result you prefer? nobody said to only continue from the first result

>>103886142
enjoy

>>103886142
No until we get a proper found footage horror prompt that doesn't light up the scene like if it was a studio

>>103886144
still gonna get worse and worse, let's suppose you find the best output everytime, let's say you go from a real image that's 100% accurate (which is the definition) and then you make a video of it, and the last frame is from the best output ever from a model, let's say it gets a 90% accuracy, then you'll have to use this 90% accuracy frame and extend it with i2v, again, you lose another 10% of accuracy, so now it's a 0.9x0.9 = 81% accuracy last frame, do you see where this is going anon?

>>103886097
We don't feed bats and mice to people and lock people in cages and expect people to serve sewer garbage to people just to stay alive, we make them go through government programs that take 6 months to go through and then deny them medical coverage, the difference is my people suffer and your people are dead, Chang faggot

>>103881279
Great detail but too slopped

>>103886178
>and lock people in cages
australia did that during the covid, what are you talking about
https://www.nytimes.com/2021/08/20/world/australia/howard-springs-quarantine.html

>>103886177
No because we don't continue forever, we are creating scenes. With Lora's we can continue the overall look and theme for consitency and start again in the next scene, which by the way will help with accuracy in the previous scene too.

>>103886178
>your people are dead
China is the 2nd most populated country in the world, they are far from dead lol

>>103886130
I don't think you are being serious, I think you're just trying to get people riled up by pretending to be retarded.

>>103886188
You'd probably live longer in Aussie being locked in a cage anyway, everything in your island wants to kill you in some way or another. You're posting on 4chan, no one would miss you anyways.

>>103886208
then I'll ask instead
why would an img2video model not be able to turn an img in2 a video

>>103886214
>You're posting on 4chan, no one would miss you anyways.
you're also posting on 4chan, is this how you think about yourself aswell? :(

>>103886200
Their birthrate and economy is dead, they are dying faster than Japan is, WW4 fucking when

>>103886221
I'll be happy when I am dead, world is fucking cancer, I hope it all goes to shit sooner than later

>>103886219
Thank you for the backup, I was losing energy talking to people without imagination.

It's like talking to the anons that kept saying local video was not possible for many years, as if new ideas and techniques can't be possible. Who knows what people will come up to make img2video even better than we think

>>103886208
Care to explain why you can't use img2video on the last fram of a video? Or do you want to continue deflecting sir

>>103886236
>I hope it all goes to shit sooner than later
I know some people who voted for Kamala just for this reason, so that the US' downfall goes as soon as possible, that's an accelerationist concept and desu I can't blame them, sometimes a better society comes out the ashes

we need a rentry on the flux shill

>>103886276
which one? there's a lot of them, flux is a popular model after all

thanks

>>103886261
yes a lot of us are tired of the bullshit with the economy and elites, I'm not going to say much because they are always watching, but if a revo was to happen I wouldn't give a shit

>>103886287
the one that throws a tantrum everytime you use something that isn't flux

>>103886291
Have you considered just using flux then? If it's that easy to please a bro...

>>103886241
I don't have any expectations, I think it'll produce coherent motion but it's still a roll of the dice, and chaining them together is a novel idea but you're still limited to how many frames you can produce in one go, anything longer will be entirely detached from the motion that came before it and will look janky as furk

>>103886291
That could honestly be me but this time it's not. It's no secret I look down at SDXL posters as lifeforms akin to cockroaches.

>>103886290
Someone is always watching because this is a simulated universe, things like stable diffusion is just like a reflection to the code leaking in a way.

If this is a simulated universe, then there is somethign calcuating everything and keeping track of everything we do.

>>103886311
Mad that XL chuds ON TOP and fluxfags on BOTTOM

>>103886311
>It's no secret I look down at SDXL posters as lifeforms akin to cockroaches.
same

>>103886307
i do use it, but only for inpainting backgrounds

fresh thread

>>103886319
>>103886319
>>103886319
>>103886319

fresh thread

>>103886219
>>103886259
You could, but the image itself contains no information about the frames before itself, so it conveys nothing about movement/etc in the video. It will create extremely jarring one-frame transitions. This will make for very bad videos.

>>103886328
I tried to explain him that but he doesn't get it, I guess he'll get it the hard way when the model will be out and he'll realize how impossible is utopia actually is

Is there any competition locally that is trained to compete with Hunyuan Video?
I see so many closed online only services like klingai, pixverse, hailuoai, (and they all have image to video) but nothing local.

>>103887110
>Is there any competition locally that is trained to compete with Hunyuan Video?
nothing is gonna top hunyuan in our lifetime, it's the first and probably last fully uncensored model, how can you beat something as amazing as that?

>>103886053
>continue any single video from using the last frame of the video
is there any work on working on the x last frames instead to at least give an idea of what's going on to the model?

>>103881746
>artist knowledge
>neutered models
Doubt it bro. SD1.5 still king for that.