[a / b / c / d / e / f / g / gif / h / hr / k / m / o / p / 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
/gif/ - Adult GIF
Worksafe Board: /wsg/


Thread archived.
You cannot reply anymore.


[Advertise on 4chan]


File: 1786749085736335.webm (3.7 MB, 1200x842)
3.7 MB
3.7 MB WEBM
Thread for AI generated videos.

Thread Rules:
>Only videos made in Local AI video generators are allowed such as WAN or LTX or others
>No videos made in Popvid or Grok or any other similar low-quality generators
>No AI Celebs
>No Prominent AI glitches for no reason
>No Blacked

>Wan 2.2 Guide:
https://rentry.org/wan22ldgguide

Previous: >>31034728
>>
File: 1786749943941258.webm (3.7 MB, 676x1200)
3.7 MB
3.7 MB WEBM
>>
File: 1786753585835963.mp4 (2.52 MB, 576x992)
2.52 MB
2.52 MB MP4
>>
File: Test 00009-Audio(10).mp4 (3.68 MB, 640x954)
3.68 MB
3.68 MB MP4
>>
>>31043824
>Wan2.2 guide
Time to remove this superseded ancient tech
And for anyone wondering “how do you prompt?” the only right answer is LLM. You have to stop being a retard, educate yourself, read the instructions and use an llm. It’s simple, it’s free, it’s fast, it’s effective, Welcome to 2026.
>>
File: MiniMax_H3_00184_.webm (3.71 MB, 1024x1024)
3.71 MB
3.71 MB WEBM
>>
>>31043866
With LTX you can still get by somewhat well with prompting yourself. In order to use Minimax H3 to its full potential, you'll probably have to use an llm.

I'm also sure the WAN guide does more harm than good these days. Most people would be better served by just installing ComfyUI Desktop, instead of cloning a repo and running some script.
>>
>>31043824
https://wixmp-ed30a86b8c4ca887773594c2.wixmp.com/v/mp4/7b42947b-d136-44dc-8c65-9e8b0a3536f2/dmmt0xt-81c11f1c-6498-44ce-99af-149783cda014.VideoQualities.res_480p.a60da46f7f3d4ca09a4acc212fc094a7.mp4
Test
>>
anyone kindly teach me how to prompt h3. i'm new to this and keep getting blocked in reference images
>>
Was using Wan2GP just fine.
But for no reasons, it suddenly nuked itself and the folder is gone.
No idea what happened. Can't be windows defender either, because I have disabled that.
>>
>>31044025
It was me, I pranked you.
>>
>>31044025
I updated wan2GP the other day and then it was incompatible with pinokio. It said pinokio needed to update but any attempt to do so just opened the one already on my pc. So I wanted to do a clean install and remove it from my pc first. When I wanted to backup my loras and such I noticed they'd already all been removed by the prior wan update. (Or the pinokio update attempts)
>>
File: mm_h3_fuck_yeah.webm (3.75 MB, 768x1376)
3.75 MB
3.75 MB WEBM
h3 is incredible
>>
>>31043866
It's all placebo, LLMs, yourself, negative prompts, doesn't matter.
>>
File: tmphg0_anqi.mp4 (1.62 MB, 832x480)
1.62 MB
1.62 MB MP4
Need a bit of help. I want to do fast paced titfucks with img2vid, but they usually go like this which are still good but not what I want. I tried it in both Wan and Seedance, and the most accurate was back when Grok used to be good.

Any good prompts to make girls titfuck crazy style?
>>
File: nemona2.webm (3.72 MB, 794x1160)
3.72 MB
3.72 MB WEBM
>>
Why did the previous thread archive at 30 posts?
>>
How do you physically link a video as reference in H3? Load video->get video components-> image output into load video input? Do i have to preprocess it to match output aspect ratio or final resolution or anything else other than length? Does using output resolution instead of full 2048px for processing the references lower its quality? Because the full one basically doubles inference time and I haven’t noticed any difference at all when generating at 1 megapixel (1344*768)
>>
File: minimax-h3_test-6step.mp4 (3.44 MB, 800x1184)
3.44 MB
3.44 MB MP4
>>
File: BRIDALMASK.mp4 (3.09 MB, 704x1056)
3.09 MB
3.09 MB MP4
>>
>>
>>31044256

It rolled past p5 without any activity
>>
File: deez nutz.webm (1.83 MB, 720x1280)
1.83 MB
1.83 MB WEBM
Imagine if this was your daughter, girlfriend, wife, mum, aunty or friend.
Fucking disgusting creeps creating fake vids/pics of random girls
Just put yourself in their shoes, or imagine it's your family or friend.
Stop doing this shit.
If I see one of you AI my daughter IRL, I'll tie you up and get my Turkish kangal to eat your nuts.
>>
File: barking daughter.mp4 (1.44 MB, 600x900)
1.44 MB
1.44 MB MP4
>>31044732
my daughter will bite your dog
>>
File: ButtsAndButts.mp4 (1.48 MB, 640x1152)
1.48 MB
1.48 MB MP4
How do you guys generate images these days? I've really stuck to using Z image turbo (cyberrealistic v7.0 https://civitai.red/models/2218365/cyberrealistic-z-image-turbo?modelVersionId=3101446)

I've seen a lot of mention of Krea 2, are there similar NSFW version of it you guys recommend?
>>
>>31044878
Krea 2 or flux klein with LoRas
>>
>>31044894
Is that just base krea 2 or turbo, or some other variant? Z image currently gives me 968x1728 in about 8 seconds per image. Any idea if Krea is similar?
>>
>>31044459
beautiful, more?
>>
Looks like somebody already posted a workflow for auto-prompt generation with H3 for the lazy/people that cannot type quickly/nonautistic folks.

https://civitai.red/models/2860715/minimax-reference-to-video-workflow-reference-manager-node-auto-prompting?modelVersionId=3231456

https://github.com/Hearmeman24/ComfyUI-MiniMaxRefPack
>>
>>31044906
you definitely need either a trained checkpoint like GonzaLomo or a LoRa and if you can't really decide, just browse civitai red images section and look up on the generation data what people are using
>>
File: Test 00027-Audio(4).mp4 (2.63 MB, 600x900)
2.63 MB
2.63 MB MP4
>>31044755
Tried to remaster this old gen with Krea2 and Minimax H3. Barking wasn't really to my liking desu

>>31044878
Krea2 + Realism Engine lora works great.You can grab a solid workflow from the preview images in civitai. Uses some WAN2.1 VAE, but it does look pretty good. Much more natural than what I remember back from when I was using Z-Image.
>>
File: 1786640831379655.mp4 (2.54 MB, 640x640)
2.54 MB
2.54 MB MP4
Guys, we need more Black Women in these threads. I've seen White, Latina and Asian (and that abomination called Futa) but hardly any Black Women. We need gens of beautiful Black women like the one I posted. And we need gens of beautiful Black Women being ravished by BBC in 3somes & gangbangs. I just know a few of you cool ones won't let me down. Thanx in advance.
>>
>>31045112
Its cuz they're ugly
>>
File: 1786471309656955.mp4 (1.31 MB, 640x640)
1.31 MB
1.31 MB MP4
>>31045159

The ones I'm posting were posted in previous threads. Obviously someone agrees with me.

Another thing. You sound like a virgin who would be a killjoy if we went to the club together. All pussy is the same color in the dark.
>>
>>31045112
Just realized Anima can't make a proper black african woman. Shame.
>>
>>31045233
Nope black girls are the ugliest, okay maybe a lil above indians but still ugly af.
>>
File: 1786409729287596.mp4 (2.36 MB, 832x1088)
2.36 MB
2.36 MB MP4
>>31045324


Coo story bro.
>>
File: 1786484959095724.mp4 (3.43 MB, 832x1216)
3.43 MB
3.43 MB MP4
>>31043824
>>
>>31045375
Very good!
>>
File: yuna ego fucked.mp4 (1.34 MB, 528x720)
1.34 MB
1.34 MB MP4
>>
File: mm_h3_fuck_yeah2.webm (3.78 MB, 768x1376)
3.78 MB
3.78 MB WEBM
>>31044058
>>
File: 1786744819741797.webm (3.69 MB, 1200x944)
3.69 MB
3.69 MB WEBM
>>31043824
I am once again requesting more gens of Eva girls, which are truly the best gens of them all
>>
File: 0515_00001.webm (3.72 MB, 1086x1200)
3.72 MB
3.72 MB WEBM
>>
File: Test 00004-Audio(1).mp4 (3.13 MB, 1024x1536)
3.13 MB
3.13 MB MP4
>>31045409
As you wish
>>
File: Yuna.webm (3.78 MB, 1080x1380)
3.78 MB
3.78 MB WEBM
>>
Did you know, the X-Men being based of civil rights leaders is a complete myth but you can still see niggers get triggered about most of them being white.
>>
File: E6G.webm (3.71 MB, 1152x768)
3.71 MB
3.71 MB WEBM
>>
>>31045605
what
>>
>>31045777
Yeah apparently all of them believe that Magneto is Malcolm X and Xavier is MLK when actually Chris Claremont based Magneto on Israel prime minister and former terrorist Menachem Begin and Xavier was based on the first Israel prime minister David Ben-Gurion

What a bunch of stupid niggers, right?
>>
File: 00125_.mp4 (3.22 MB, 640x1024)
3.22 MB
3.22 MB MP4
>>31045852

If we are just posting bullshit nonsense.. here ya go
>>
I don’t get who would use crypto to pay for nsfw buzz on civai when its all publicly traceable now? Probably even less “private” than using credit card. Unless you use really sketchy means.
>>
>>31045324
Generally true
>>
>>31043920
I dont yet trust an llm with the prompt I think it will mess up details but maybe I am wrong… h3 does take a lot more prompt leg work but once done it seems to mostly obey
>>
>>31046047
>h3 does take a lot more prompt leg work but once done it seems to mostly obey
Could you post some examples?
>>
File: 1786724603010120.webm (3.82 MB, 1680x952)
3.82 MB
3.82 MB WEBM
>>31046144
NTA - I don't think I've done a single H3 prompt that would fit into the 2,000 character limit. That's why I made this, which shows the ref2va prompt that was used to create it.
>>
>>31046047
You don't have to fully trust an LLM. Just let it do the annoying parts. Describe the reference images, how they relate to the subjects, and how the shots are generally composed. No reason you can't edit that after an LLM is done with it.

Prompt adherence really is great, especially compared to stuff like LTX.
>>
>>31046160
Do you do anything with fast motion like action scenes? Particularly with live action I tend to get a lot of blur and face distortion. I feel like most stuff posted here is single scenes without a lot of fast motion.
>>
File: 1786340579942433.webm (3.59 MB, 693x400)
3.59 MB
3.59 MB WEBM
>>31046199
Faster scenes with a lot of motion tend to need more steps, higher resolutions, and unambiguous prompting. I've had decent success but I've been avoiding it because of my hardware limitations. 10 minute gens don't bother me, but I only have to push it a little bit to turn those into 1hr+, so getting good coherent fast motion is prohibitive for me personally.
>>
File: MiniMax_H3_00011_.webm (3.9 MB, 1056x608)
3.9 MB
3.9 MB WEBM
>>31046230
You probably have a better setup than me. I only have a 3060 with 12gb of vram and 64 gb ram. Do you use any speedups or turbo loras, how many steps do you usually run? I did this text to video with the 8 step turbo lora at .6 megapixels. Took about 25 minutes. If I could find settings that gave decent results, I wouldnt mind waiting if I had something decent I wanted to gen. Unfortunately testing is slow with my rig. I might try a shorter gen at higher step count to see if that helps.
>>
File: 1786586819605810.webm (3.77 MB, 1088x832)
3.77 MB
3.77 MB WEBM
>>31046259
I have a 3070 with 8GB VRAM and 32GB of RAM, so I don't want to hear it you 12GB-havin' rich bitch.
I've had success with lightx2v turbo loras, but I've not tried the new ref2va specific one yet. Usually I use the 8 step turbo lora, at 8 steps, 0.4-0.6 strength usually does good with 20B pruned, 0.5-1.0 with the full 33B. When I want something nice I turn off the turbo lora and do 20 steps, I've played with more but I rarely see improvement beyond 20. I also use a Qwen3-VL quant (I don't recall off hand what specifically, I think it's INT8), and an FP8 VAE.
>>
>>31046285
lol sorry, your stuff is pretty good quality so I assumed you were working with something beefier. I feel like i had a better handle on WAN but maybe that was because it was shorter, simpler gens and had so many loras to help. Hopefully I can get into a groove where I can prompt whatever I want it just works but right now I'm struggling lol.
>>
>>31046300
I always felt limited by being unable to direct WAN, I feel relieved by H3, I'm really taken with it. Follow the official prompting guides to the letter, if you haven't read them yet: https://huggingface.co/MiniMaxAI/MiniMax-H3/tree/main/docs
>>
File: MiniMax_H3_00003_.mp4 (2.76 MB, 1056x608)
2.76 MB
2.76 MB MP4
Didn't come out exactly how I wanted but I still like it lol
>>
>>31046311
this is exactly what I came here to see, you're doing God's work
>>
>>31046308
I basically come up with an idea and have Grok format it using that guide. Then do a little revision. It mostly works. Feel like I'm like 80 percent there
>>
>>31046311
wild. great job
>>
File: MiniMax_H3_00004_.mp4 (2.4 MB, 1056x608)
2.4 MB
2.4 MB MP4
>>31046315
Thanks.
>>31046319
I also sometimes tell Grok to use the same prompt but in a cartoon or anime style, which tends to look better most of the time imo
>>
>>31046336
>TTG raven suddenly on the strike screen
lol
>>
File: ohAnon.mp4 (1.24 MB, 704x1056)
1.24 MB
1.24 MB MP4
Don't mind me, just testing some voice capabilities
>>
>>31046319
I put together a skill for pi agent, so far it's worked well with Sol. Made it from summarized versions of the docs for H3 and for WanGP, so I can discuss an idea with it then it spits out the prompt and specific settings for that gen to use. It's overkill but I'm happy with it. It can also just do it itself, it can launch WanGP on its own and prompt it directly but I haven't really bothered with that yet beyond testing. Paired with a good enough multimodal model could probably task it with making some fun stuff with very little input.
>>
>>31045112
>>
>>31046396


Yes, keep'em coming. An AI thread without Black Women is like a Toy's R Us with no toys or video games.
>>
>>31044919
Seems like you need openrouter credits for this. If you turn it off, it doesn't generate a real prompt in the format and just keeps whatever caveman idea you write in.
>>
>>31045517
Got a kek out of me, needs a variant where she angrily strips nude
>>
>>31046311
Man, The Big Lebowski is different than I remembered...
>>
File: wanvideo__00009.rife.mp4 (3.76 MB, 448x800)
3.76 MB
3.76 MB MP4
>>
>>31046396
how can people think this is even remotely sexy

like, if she had more realistic proportions, sure - but shit like this only looks good in drawings, not in photorealistic video, jfc
>>
>>31046631
these videos are supposed to be sexy?
>>
>>31046311
Ben Stiller will never turn down a role
>>
>>31045590
preferable to actual mosquitoes desu
>>
File: tentacles.webm (3.74 MB, 1280x720)
3.74 MB
3.74 MB WEBM
>>
File: 0002.webm (3.73 MB, 480x832)
3.73 MB
3.73 MB WEBM
>>
File: shut_up_mutt.mp4 (2.39 MB, 1024x1536)
2.39 MB
2.39 MB MP4
>>31044732
>>
What lora are you guys using to supplement the video generation?
Problems I'm seeing now is that
>Boobs are stiff
>nipples, vaginas & dicks are fake
>fucking motion is weak, it doesn't go all the way in & out
>>
>>31046953
>Boobs are stiff
https://civitai.red/models/2856004/breast-play-and-jiggle-for-minimax-h3
>nipples, vaginas & dicks are fake
Ref2VA & https://civitai.red/models/2846342/hmpussy-pussyanus-lora-for-minimax-h3
>fucking motion is weak, it doesn't go all the way in & out
https://civitai.red/models/2856467/mmh3-mystic-xxx + https://civitai.red/models/2840146/faster-harder-shake-harder-or-h3-motion-booster
>>
>>31046966
Thanks a lot! I will check them out.
>>
File: aur-talk_test.mp4 (3.89 MB, 672x1216)
3.89 MB
3.89 MB MP4
>>
>>31046430
>>
>>31046631
Shut up, faggot!
>>
>>31047025
Me and my wife
>>
>>31046631
you wouldn't get it man
>>
File: 112.webm (3.69 MB, 1536x1536)
3.69 MB
3.69 MB WEBM
It's been a little, here are more
>>
>>31047025
your wife has a pretty big dick anon.
>>
>>31047129
Hot
>>
File: 116.webm (3.64 MB, 1536x1536)
3.64 MB
3.64 MB WEBM
>>31047129
I don't know why she runs like this, audio boy is really loud in this
>>
File: 125.webm (3.7 MB, 1536x1536)
3.7 MB
3.7 MB WEBM
>>31047140
other side
>>
File: 117.webm (3.7 MB, 1536x1536)
3.7 MB
3.7 MB WEBM
>>31047143
Couldn't think of a cool effect for x-ray vision and this was just a little silly so didn't want to work on it too much
>>
File: 118.webm (3.73 MB, 1536x1536)
3.73 MB
3.73 MB WEBM
>>31047145
here is some neat Ravenclaw lore
>>
File: 120.webm (3.72 MB, 1536x1536)
3.72 MB
3.72 MB WEBM
>>31047148
faces look derp from far away and I didn't really need a far away establishing shot
>>
>>31044025
It happened again.
I wonder what the fuck is going on that it keeps self destructing.
Or that something is deleting it.
>>
File: 145.webm (3.72 MB, 1536x1536)
3.72 MB
3.72 MB WEBM
>>31047154
this one took a embarrassing amount of time to prompt right and I still don't like it
>>
File: 122.webm (3.65 MB, 1536x1536)
3.65 MB
3.65 MB WEBM
>>31047161
this one did it in one try so I didn't even bother fixing her to look like her more
>>
File: 126.webm (3.68 MB, 1536x1536)
3.68 MB
3.68 MB WEBM
>>31047167
not going to lie this one is fucking weird
>>
>>31047161
You're such a delight, anon, thanks for sharing.
>>
>>31047158

Options to possibly solve issue:

- Search reddit/huggingface/youtube for possible solution
- Try ChatGPT
- Try Google AI

>> Comment about it on 4chan
>>
File: MiniMax_H3_00004_.mp4 (1.81 MB, 640x640)
1.81 MB
1.81 MB MP4
first time making nsfw video
resolution got fucked up a little
>>
File: 127.webm (3.71 MB, 1536x1536)
3.71 MB
3.71 MB WEBM
>>31047171
measurements for... stuff

>>31047173
https://github.com/GKartist75/wan2gp-desktop
This one is recommended by
https://github.com/deepbeepmeep/Wan2GP
(you can find it on the github if you don't want to click a random link from a anon)
>>
File: 129.webm (3.74 MB, 1536x1536)
3.74 MB
3.74 MB WEBM
>>31047183
made this just for his stupid laugh
>>
>>31047189
>>31047183
are you using minimax?
t2v or i2v?
>>
File: 134.webm (3.66 MB, 1536x1536)
3.66 MB
3.66 MB WEBM
>>31047189
*crowd noises*

>>31047194
Minimax all t2v no loras
>>
File: 131.webm (3.71 MB, 1536x1536)
3.71 MB
3.71 MB WEBM
>>31047200
This one took like 4 days of prompt fixing
It wasn't really worth it I just got so annoyed whenever this popped up from my gen list it just game out shit I just HAD to get it
>>
>>31047173
Oh wow, you must be the smartest person alive.
Because surely nobody else could thought of doing those things.
Have you considered getting the nobel prize for being the smartest little fella online?
>>
File: 132.webm (3.73 MB, 1536x1536)
3.73 MB
3.73 MB WEBM
>>31047208
audio, the student is really loud for some reason
>>
File: 136.webm (3.72 MB, 1536x1536)
3.72 MB
3.72 MB WEBM
>>31047218
kind eh on this, she doesn't get grabbed like the other one but I really like how she got spat out
>>
File: 133.webm (3.71 MB, 1536x1536)
3.71 MB
3.71 MB WEBM
>>31047223
I don't normally post dupes in the same thread but fuck it
>>
File: 135.webm (3.72 MB, 1536x1532)
3.72 MB
3.72 MB WEBM
>>31047226
another fucking weird one not going to lie
>>
File: 139.webm (3.7 MB, 1536x1536)
3.7 MB
3.7 MB WEBM
>>31047229
oops
>>
File: 149.webm (3.69 MB, 1536x1536)
3.69 MB
3.69 MB WEBM
>>31047233
>>
>>31047200
how to do you get the Hogwarts aesthetic with just t2v?
>>
File: 146.webm (3.7 MB, 1536x1534)
3.7 MB
3.7 MB WEBM
>>31047239
their robotic dialogue is funny
>>
>>31047239
great stuff, keep up the good work
>>
File: 147.webm (3.72 MB, 1536x1536)
3.72 MB
3.72 MB WEBM
>>31047245

>>31047243
It knows what the Harry Potter movies looks like and because it's a movie (I think) the voices tend to come out not super loud(usually)
>>
File: 150.webm (3.68 MB, 1536x1536)
3.68 MB
3.68 MB WEBM
>>31047251
>>
File: 151.webm (3.68 MB, 1536x1536)
3.68 MB
3.68 MB WEBM
>>31047254
>>
File: 155.webm (3.69 MB, 1536x1536)
3.69 MB
3.69 MB WEBM
>>31047257
last one, night gamers
>>
>>31047258
Good night Potteranon
>>
>>31047258
These are awesome. I want to make HP stuff myself but >students
How hard do you have to fight to keep their ages up there?
>>
File: storage.mp4 (2.38 MB, 864x576)
2.38 MB
2.38 MB MP4
>>
What llm or nodes do you guys use to generate H3 appropriate prompts is this one of those times where I'd want to use those heretic models? i'm getting tired of writing prompts i'd rather have some node deal with it as much as possible
>>
>>31047337
Just give an LLM the documentation and the rules and ask it to make prompts for you. Grok can do NSFW straight out, or just use other sfw phrases/words in place of what you want and then just edit their output to your liking.
>>
File: Quit_struggling.mp4 (1.44 MB, 1536x1024)
1.44 MB
1.44 MB MP4
>>
File: ref_MiniMax_H3_00015_.mp4 (985 KB, 1376x768)
985 KB
985 KB MP4
>>31016825
proof of concept for removing tattoos (and watermarks)
>>
>>31047210

>> comes to 4chan to bitch

Nobody fuckin cares loser
>>
>>31047384
Fun idea
>>
File: 1784966433267.webm (3.9 MB, 640x360)
3.9 MB
3.9 MB WEBM
>>31043824
I'm reposting this vid from the old thread as I was impressed with the coherence, length and quality. I'm making a system to stitch vids like this together for longer material too.
>>
>>31047161
More vore!
>>
Once again asking for weird sexual gens, eg mouth nipples, displaced pussies, dicknecks, etc
>>
Hello i'm looking for some gens of my irls, please add me on teleguard

L8S2CXGD7
>>
>>31047480
this guy is a fed btw
>>
File: h3_crotchface2.webm (3.2 MB, 544x960)
3.2 MB
3.2 MB WEBM
>>31047464
How's this
>>
>>31047599
Hell yeah! That's the stuff
>>
>>31047480
Too bad the only guy who would be up to teach you how probably hates all of this.
>>
>>31047480
Your iris? You making sharingan videos dude?
>>
>>31047428
amazing
>>
File: magictrick1.mp4 (2.2 MB, 864x576)
2.2 MB
2.2 MB MP4
>>
>>31047384
KILL ALL TATTOO ARTISTS

THANK YOU FOR YOUR SERVICE
>>
>>31043902
She looked better before
>>
H3 is stressing me out
Why i even bother. Fuck their prompt. Fuck chinks and their soulless way of thinking
>>
>>31047599
Would
>>
>>31047844
Prompt too fucking long
Character speak gibberish. "No dialogue, No talking, No grunting" dont even work

FUCK MINIMAX
Time to go back to LTX
>>
>>31047384
insanely based!
>>
>>31047858
>Character speak gibberish. "No dialogue, No talking, No grunting" dont even work
this is my issue too you have to be very aggressive, but the output is generally better than LTX
>>
>>31047880
Bro i tried
E V E R Y T H I N G
8 HOURS
HUNDREDS OF FAILED GENS
ZERO GOOD GEN
FUCK THAT
>>
I'm impressed by the potter content. Would a 5080 be enough to gen that shit?
>>
>>31047908
A 4070 Super is enough
>>
>>31047226
>>31047229
Those worked out pretty well. First chest was decent but second one worked out even better. The pigmen were funny.
>>
File: 1786864652.webm (324 KB, 1344x1216)
324 KB
324 KB WEBM
>>
>>31047428
Fucking perfect!
>>
File: 1773611530905309.mp4 (1.05 MB, 1184x896)
1.05 MB
1.05 MB MP4
>>
File: test.webm (3.29 MB, 580x1040)
3.29 MB
3.29 MB WEBM
>>
subject_definitions:
<Subject 1> is the 20yo woman with ponytail in <Picture 1>.
<Subject 2> is the nude male standing still on left screen in <Picture 1>.
<Subject 3> is the nude male on right screen in <Picture 1>.
<Subject 4> is the nude male wearing sunglasses laying down on large table in <Picture 1>.
<Subject 5> is the nude male on top of woman.

summary:
[reference generation] The target video features <Subject 1> engaging in sexual activity with four males. The scene utilizes the visual style and character design from <Picture 1>. The audio references the moans from <Audio 1> for the woman while the males remain silent.

retention_analysis:
<Subject 1> (appears in [Shot 1]): partially_preserved - the woman s hairstyle, and facial features are based on <Picture 2>.
<Subject 2> (appears in [Shot 1]): fully_preserved - the nude male is retained.
<Subject 3> (appears in [Shot 1]): fully_preserved - the nude male is retained.
<Subject 4> (appears in [Shot 1]): fully_preserved - the male wearing sunglasses is retained.
<Subject 5> (appears in [Shot 1]): fully_preserved - the nude male on top of woman is retained.
<Audio 1>: reference - the voice timbre and moaning sounds referenced from <Audio 1> guide the dialogue delivery of <Subject 1> without copying the original signal.

detailed_description: [Shot 1] <Subject 1> strokes <Subject 2> hard penis while <Subject 3> and <Subject 4> moving their hips back and forth in turn at high speed to bump <Subject 1> crotch, and <Subject 1> anus is stretching back and forth as <Subject 3> and <Subject 4> hard penis penetrate <Subject 1> anus. <Subject 3> is holding a blue stick firmly with right hand and moves his hand up and down nonstop while blue stick penetrate <Subject 1> vagina.
<Subject 5> slides his hips and hard penis back and forth at high speed towards <Subject 1>, breasts chest and <Subject 5> crotch slightly hits <Subject 1>.

overall_soundscape:
N/A

non_diegetic_music:
N/A

---
And its still fail lol, im done.
>>
>>31048020
What's your GPU,CPU, and RAM?
>>
>>31048032
RTX 5090, 96gb of ram.
Im not having problem with the motion.
Im having problem with the AUDIO.
The fucking males make annoying "ngggggghh" "ggrrrrrrkkkkk" sound and its boner killing
>>
I'm sad that I'm not creative enough to think of scenarios to prompt about.
Taking off clothes & grabbing boobs seems to be the only things I know to prompt about
>>
File: BRIDALMASK.mp4 (2.74 MB, 704x1056)
2.74 MB
2.74 MB MP4
>>
>>31048084
kino
>>
>>31047691
>>31047314
Is there more of these?
>>
>>31048127
We get what we get.
>>
File: MiniMax_H3_00046_.mp4 (2.82 MB, 960x544)
2.82 MB
2.82 MB MP4
>>31047223
>>31047226
very nice, tried it myself but without it just being implied, not great but hopefully we can get an actual lora for this kind of thing so it will rival JAVs
>>
File: HE.mp4 (1.79 MB, 576x896)
1.79 MB
1.79 MB MP4
>>
File: astolfo.webm (3.7 MB, 960x1440)
3.7 MB
3.7 MB WEBM
>>
>trying to make a bra undressing video
>shit is phasing around all over the place
Fuck!
>>
I just wasting my sunday for this
>>
File: Gi5.mp4 (3.43 MB, 608x896)
3.43 MB
3.43 MB MP4
>>
>>31048360
probably easiest to make her unclasp it from the middle bridge part
>>
>>31048372
Very good!
One detail I would like is to give her slutty makeup only after she changes.

>>31048377
Yeah, probably.
>>
>>31048360
Post your prompt. Minimax H3 requires specific format, free text will most likely not work very well.
>>
Good thing about newer models it’s their darwinism where dumb retards get cut out more and more
>>
>>31048412
I've had LLM generate it.

https://sharetext.io/u456wcod
>>
>>31048360
Reminder, try all prompts without loras.
A lot of the lora on civi at 1.0 strength absolutely destroy prompt adherence and introduce the jankiest movements
>>
File: bra strip compressed.mp4 (3.78 MB, 896x1184)
3.78 MB
3.78 MB MP4
>>31048446
Best I could get.
>>
Try boomer prompting
i.e

<subject 1> is the hot grill in the reference image, her bra is taken off, teasing for the camera, multiple camera angles and pans
>>
>>31048446
What the fuck is this prompt? What broken ass lllm did you use to generate such shit. It’s not even properly structured. The overall_soundscape is a masterpiece
>>
>>31048462
GLM 5.2

But I guess I have to do shit manually.
>>
File: MiniMax_H3_00057_.mp4 (414 KB, 608x1056)
414 KB
414 KB MP4
>>31048446
Well, I cant view it, the server is crapping out, but other people seem to not like it.

This is what I am using for creating characters that I am going to use later:
subject_definitions:
<Subject 1> is the woman from <Picture 1>, <Picture 2>, <Picture 3>. <Subject 1> tribal clothes and tiara with hawk feathers. <Subject 1> has tattoos

summary:
The target video presents <Subject 1> in various poses and from various angles. <Subject 1> is standing in a featureless black void bathed in gentle uniform natural light. The target video should not contain any background, characters or items from the provided references.

retention_analysis:
<Subject 1> (appears in [Shot 1], [Shot 2], [Shot 3], [Shot 4]): fully_preserved - the characters body and visual appearance are preserved.

detailed_description:
Live action, full shot of <Subject 1>. <Subject 1> is assuming neutral standing pose.

[Shot 1] <Subject 1> is standing facing forward.
[Shot 2] At 00:00.500 quick cut <Subject 1> is standing facing left
[Shot 3] At 00:01.000 quick cut <Subject 1> is standing facing back
[Shot 4] At 00:01.500 quick cut <Subject 1> is standing facing right

overall_soundscape: N/A

non_diegetic_music: N/A
>>
File: MiniMax_H3_00065_.mp4 (329 KB, 672x1216)
329 KB
329 KB MP4
>>31048484
This was the result for the prompt btw - pretty good given the reference images were not perfect.

The only thing is the sound - it seems NA does not work as it does with music, but it does not bother me here, this video is for reference, not the finished product.
>>
File: upscaled_00006_1.webm (3.69 MB, 1600x1152)
3.69 MB
3.69 MB WEBM
>>
File: Gi6.mp4 (1.55 MB, 608x896)
1.55 MB
1.55 MB MP4
>>
>>31048651
fucking lel
>>
>>31048077
Nothing wrong with classic nudity and titty/ass groping. Just use different characters/clothes/locations/poses/styles/intensity/camera angles/music etc.

You can still get a lot of variation out of those basic concepts
>>
I'm out of the loop, is minimax the new wan for local gens and is there a guide for retards like the wan rentry?
>>
>>31048244
to old
>>
>>31048651
Laughed ngl
>>
>>31045233
>this guy only fucks in the dark
Imagine not wanting some lights on during sex because the woman is so beautiful.
>>
>>31047599
Face crotch atm when?
>>
File: 1786393820065225.mp4 (1.51 MB, 448x672)
1.51 MB
1.51 MB MP4
To the anon who gened this one: keep them coming! Simple but great.
>>
>>31048358
Yes! More hung femboys!
>>
Be careful what you ponder.
>>
Anyone else feeling drunk or getting addicted to AI generation? I can't even begin to explain how incredible it is that we're now able to do this. With just a personal computer and a decent enough vocabulary and capability to illustrate your thoughts into words, a person basically has the closest thing to reality-warping powers. It feels unreal to have such an insane amount of power and control at such a bargain.
>>
>>31049015
You'll get over it.
>>
>>31049015
>decent enough vocabulary and capability to illustrate your thoughts into words
Yeah. Rest of the fucking owl.
>>
>>31049015
No, it's super tedious and annoying to keep rolling the prompt gatcha over and over again. It's never exactly how I want it, and it fights me at every step.
>>
>>31047200
Any more clothes ripping off vids? Or embarrassed nude women?
>>
File: TEST35.mp4 (1015 KB, 512x800)
1015 KB
1015 KB MP4
>>31048484
I think I might find uses for this kind of prompt, thanks anon
>>
How do I use multiple loras on the Wan2GP?
Seems like the video gets blurry & smeary when I use more than 1 loras.
>>
>>31049015
And the results always look like shit, literally zero soul, and no effort or love goes into the creation. Every piece of AI generated video is worth infinitely less than traditional content. You have AI psychosis. Log off and get a hobby
>>
>>31049015
Only if your prompt works.
Lots of people got filtered with Minimax
>>
>>31048020

Did ya try something easier first, say a solo scene or maybe just 2 people instead of 5. Trying a 1st gen with 5 subjects with complicated motions as a 1st gen is like trying quantum physics before basic algebra.
>>
>>31049015
Yeah, it's pretty much like having a lucid dream. And it will only get better (or worse, depending on the point of view), as it advances, control over generation will improve and soon you'll be able to turn any thoughts into video
>>
>>31049015

He's still in the gambler addict phase where its fun seeing what random crap the AI spits out each time.
>>
>>31049092
Already working with R2V models. I guess I2V models is just too much for gangbangs like that. It just R2V kills the motion. We need better turbo lroa so bad
>>
>>31049088
key is to input references not words
applies to both video and image gen
>>
File: 659970980268326.mp4 (3.77 MB, 544x960)
3.77 MB
3.77 MB MP4
>>
So, this "Single-Node" trend is happening right now across different models for ComfyUI. It's where everything is an all-in-one single node and you just slap it down and go.
MiniMax-H3 just got its first one from a redditor if you care to check it out.
>>
>>31048039
>Doesn't define soundscape at all
>Mad about random soundscape that emerges because he didn't bother to define it
They tell you in the documentation that this is what will happen, but you couldn't bother to read it.
>>
>>31049190

So this internet trend is happening right now you just go click click click and go. Apple gots it from a redditor if you care to check it out.

Holy vagueness batman lol
>>
>>31048039

>>31046160
>>
>>31049251
It doesnt work well with I2V models
>>
>>31047229
>They breed quickly down there in the dark, but perhaps we can slay them even faster
>>
>>31049288
>It doesnt work well with I2V models
Bullshit. MiniMax H3, both FL2VA and Ref2VA, both take direction extremely well when you actually give it regardless of what workflow you're doing. That guy left his soundscape totally undefined and then complained when it didn't come out silent - if he wanted it silent he needed to actually specify.
>>
>>31049312
>Bullshit
I tried it myself for 12 hours straight. Its no bullshit, Some poses dont work with I2V models. Same prompt. Same shit. R2V works better
>>
>>31049319
What does this have to do with poses? Did you reply to the wrong post?
>>
File: MiniMax_H3_00130_.mp4 (2.32 MB, 864x480)
2.32 MB
2.32 MB MP4
minimax 3 - working with simple prompts and references. Two pics used to obtain result.
>>
>>31049312
Also what im doing is anime styled model. Just try it yourself. They forgot to remove the speech training on the anime. Same with Low framerate DBZ tier animation
>>31049322
Its me. I2V dont work well with some poses. You know what pose mean right ?
>>
>>31049326
Not in this context, not sure what you're referring to in terms of audio and speech direction.
>>
>>31049329
What i am saying is. Minimax is bad for Anime, especially if its for porn
>>
>>31049329
Because in anime. The voice actor carries the medium. They carried it on AI models as well. This is why you always hear gibberish talking on most anime poses

What we need is a LORA to remove said speech. Because its impossible to do it in base Minimax model
>>
File: 2580921.mp4 (3.94 MB, 832x640)
3.94 MB
3.94 MB MP4
>>31048039
>>31049251
Yeah, like this anon said you need to define it. It doesn't have to be very specific though. In this video for example:

overall_soundscape:
kissing sounds, licking sounds, and muffled moans.

non_diegetic_music: N/A

>>31049190
Sounds interesting. I was never a fan of the node based ui, but I got used to it. This could be good for people that don't want to bother with the noodles.
>>
>>31049335
I agree, by the way, it's generally bad at anime and the sub-framerate shit is garbage. I'd love to see the prompt you're using that produces gibberish against your will - because I've still never experienced what you're talking about except when it was an issue with the prompt. I don't mean it to dismiss what you're saying, I want to see the problem and address it.
>>
>>31049338
Now try with anime. I guaranteed it you will get gibberish speech and groans for like 70% of the time.

>>31049342
Just try it with anime and post it in here dude.
Just generate [nsfw, 1girl, 1boys, sex ]
>>
File: 69d3b50402816b263d32300f.mp4 (3.71 MB, 1112x834)
3.71 MB
3.71 MB MP4
>>
File: 6a32aef450057a1c442e2d0b.mp4 (3.72 MB, 720x1280)
3.72 MB
3.72 MB MP4
>>31049361
>>
>>31049361
why the fuck is this video doubled up as if I held it up in front of my eyes
>>
File: 69d3b78e02816b263d345c03.mp4 (3.72 MB, 834x1112)
3.72 MB
3.72 MB MP4
>>31049366
>>
>>31049369

Why the fuck does 4chan have a 4mb limit. Compression dude, deal
>>
>>31044518
TWO
>>
>>31049352
I'll bust out some anime with no speech for you. While that's cooking, I mean it, you should share your prompt. This is communal, nigga.
>>
>>31049402

>>31048020
>>
>>31049402
Wait heres a simpler one. Still making gibberish speech when i want is just dialogueless giggling
-----------
subject_definitions: <Subject 1> is 20 year old long haired blonde girl from <Picture 1>. <Audio 1> is the voice-timbre reference for <Subject 1> and [Shot 1].
<Subject 2> : nude large out of frame male on right screen.

retention_analysis:
attribute_transfer: <Picture 1> ([Shot 1] first frame): fully_preserved.
<Subject 1> (appears in [Shot 1]): fully_preserved.
<Subject 2> (appears in [Shot 1]): fully_preserved.
<Audio 1>: reference - the target speaker follows <Audio 1>'s voice timbre and measured delivery without copying the original signal.

Animation_Style : 3d animation.

detailed_description:

[Shot 1] at 0.00 seconds : <Subject 1> use her right hand to actively stroke the large, hard erect penis <Subject 2> non stop, while holding the tip of the <Subject 2> penis with her left hand firmly. while <Subject 1> eyes is always locked at out of frame nude male on the right screen, <Subject 1> head tilting left to right sensually while making a sensual body movement. maintain <Subject 1> face expression while looking right.

overall_soundscape: <Subject 1> : faint giggling and smug breathing.

non_diegetic_music: N/A.

camera : static floaty camera
>>
>>31049408
Yeah you didn't define the soundscape, that will produce gibberish by default.
>>
>>31049424
I did it after that hours ago.
overall_soundscape:
<Subject 1> only do calm slow moaning and breathing noise.
<Subject 2> remains silent, Wet skin slaps on penis.
<Subject 3> remains silent, Wet skin slaps on penis.
<Subject 4> remains silent, Wet skin slaps on penis.
<Subject 5> remains silent, Wet skin slaps on penis.

Still failed on I2V. The males still make groaning sound.
On R2V theres no problem....... except the motion is shit on R2V models
Theres a bug on I2V models, im sure of it.

Why im using R2V nodes you ask ?? The reference sound. Its really good. But its buggy as hell.
>>
File: 980566389457925.mp4 (3.98 MB, 864x608)
3.98 MB
3.98 MB MP4
>>31049352
Idk, maybe I'm just built different, but it hasn't been a problem on my end. Even tried a few different styles just in case.
>>
>>31049419
reference audio is too long? the limit is 15s
>>
File: shower.webm (3.59 MB, 640x832)
3.59 MB
3.59 MB WEBM
>>31049419
Again, you didn't define the sounds. Your soundscape there, with no other audio direction, is just going to make all characters on screen randomly giggle and possibly produce gibberish.
>For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.
>integrated_multimodal_description: [Shot 1] The shot begins exactly from <Picture 1>, preserving the adult man and
>woman’s appearances, black-and-white anime doujin linework, screentones, shower setting, embrace, and original
>framing. The camera remains completely static. Across five seconds, they stay standing in the same passionate,
>dramatic embrace with only barely perceptible breathing and a subtle tightening of their hold. Shower water streams
>continuously over them as droplets trace downward. Neither person speaks or moves their lips. End with them still
>embraced beneath the running shower, the composition virtually unchanged.
>overall_soundscape: Only steady shower spray and soft water splashing against the tiled surfaces. Absolutely no
>speech, whispering, vocalizations, or other human mouth sounds.
>non_diegetic_music: N/A
>>
>>31049436
Dont make action that involves their mouth. Try a simple handjob with Males visible.

>>31049440
Read >>31049433
>>
>>31049440
Just try do a simple handjob. Male face visible. Not involving mouth.
>>
File: LK3.mp4 (3.87 MB, 608x896)
3.87 MB
3.87 MB MP4
>>
File: 465784461478811.mp4 (3.91 MB, 864x608)
3.91 MB
3.91 MB MP4
>>31049449
Okay, I'll try one where the man's face is visible.
>>
>>31049419

Plot twist, <Audio 1> reference is just a dude grunting LOL
>>
File: 1761937974924528.webm (3.22 MB, 960x1440)
3.22 MB
3.22 MB WEBM
Minimax problem with gibberish speech. Currently no solution
>>
>>31049498
that's just japanese you lacist
>>
>>31049015
It couldn’t have come soon enough, the catalog on this board shows the state of modern porn.
>>
File: 1782955523030903.webm (3.34 MB, 960x1440)
3.34 MB
3.34 MB WEBM
>>31049498
.....except when you adding a proper dialogue. But i just want a non dialogue giggling and laughing dammit.
>>
>>31047384
MORE famous tattoo removal ones
>>
>>31049131
Hooooly fuck best itt. Evie > eve
>>
File: 1005592771824171.mp4 (3.51 MB, 608x864)
3.51 MB
3.51 MB MP4
>>31049449
Idk what to tell you man, for me it just werks.
>>
>>31049015
>bargain

The 5090 is over 3k right now.
The new 6000 model is literally almost 15k.
>>
File: 5592771824171.mp4 (3.6 MB, 608x864)
3.6 MB
3.6 MB MP4
>>31049548
Ah shit, uploaded it without audio.
>>
>>31049552

Think he means that all these models and loras are free when they could easily have charged us for them.. and we probably would have paid up.
>>
>>31049555
What are your prompt ? I2V model ?
>>
File: MiniMax_H3__00056_.mp4 (1.17 MB, 736x576)
1.17 MB
1.17 MB MP4
>>31049419
I also get her talking gibberish
>>
>>31047428
saved. That's something.
>>
>>31048997
Would have been funnier if it had been a femboy and the pantyface went from horrified to slow acceptance.
>>
>>31047428
fuck, i forgot how good this song was
>>
File: 1092763497486524.mp4 (3.9 MB, 608x864)
3.9 MB
3.9 MB MP4
>>31049562
Yeah, fl2v model.

Asuka Langley, a beautiful young girl with red hair and soft blue eyes. Asuka is giving a handjob to the erect penis of Kaji, the blindfolded man tied to the cross. She is sliding her hand up and down his erect penis. Both Asuka and Kaji remain completely silent. Asuka hits Kaji with the whip one time, the whip makes a loud crack sound. Micro push-in + locked-off hold: Begin completely still, then make an almost imperceptible push toward the subject as their expression changes, minimal composition, long contemplative take.

overall_soundscape:
handjob sounds, whip crack sound, ambient room sounds.

non_diegetic_music: N/A
>>
File: smug.webm (3.48 MB, 512x768)
3.48 MB
3.48 MB WEBM
>>31049449
>>31049498
Assign (S1) only to <Subject 1>, bind <Audio 1> exclusively to her, give <Subject 2> no speaker ID, anchor the giggle in the timeline, not only the soundscape. I googled "male anime face" and I don't regret it.

For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.
integrated_multimodal_description: [Shot 1] The anime-style shot begins exactly from <Picture 1>, preserving the
clearly adult woman and man’s identities, visible appearances, poses, screen positions, bedroom layout, lighting, and
original composition. The camera remains static. The woman stays seated across from the standing man, watching him
with the same smug expression. He remains upright and essentially motionless.
She strokes his penis gently with her left hand, while gently and slightly rubbing the tip with the fingertips of her left hand.
Midway through the shot, the seated woman (S1) gives one brief, soft, clearly wordless giggle while keeping her gaze
on the man. The giggle contains no spoken syllables, and her lips never articulate words. The man produces no voice or
audible reaction, he stares excitedly and remains still.
End with her still seated and looking smugly at him while he remains standing, with the framing
nearly unchanged.
overall_soundscape: One brief wordless giggle from the seated woman (S1) only, clearly localized to her position.
Otherwise complete silence: no bedroom ambience, room tone, dialogue, whispers, gibberish, or vocal sound from the
man.
non_diegetic_music: N/A.

>>31049552
You only need 8GB VRAM and 32GB of system RAM to run full the full 33B, you're limited in single-gen duration and resolution and it'll be much slower than something more capable, but the barrier to entry isn't that high.
>>
Besides needing more disk space, how would you guys rate this as a AI video generation rig?

https://www.newegg.ca/asus-ascent-gx10-mini-pc/p/N82E16859110044?item=N82E16859110044
>>
>>31049626
Decent, expensive. It'll give you the room for long, high res gens, but at speeds that compete with a 3060. You would probably be able to do longer, high res gens at a similar speed with just a 3060Ti and 64GB of DDR5, which will cost a fraction of that thing. For that price you could do much, much better for videogen, those things are really best suited to running MoE LLM models.
>>
>>31049626
This is more aimed at people who want to run LLMs.
>>
>>31049015
actually, i feel like i was more addicted when it was worse, because a bad result happened more often and you'd have to re-roll. Now, it's only a couple until it's half-way decent, and when it is fucked up, it's pretty much entirely my fault for not prompting clearly enough.
>>
File: 1772645080843858.webm (2.85 MB, 1056x1584)
2.85 MB
2.85 MB WEBM
>>31049601
>>31049596
I just really dont get it. This is my monkey tier prompt without following the guide. Its hit and miss. maybe i should follow the guide at all and just use the instinct. Prompting with Minimax is really a mess.
>>
File: 614788.mp4 (3.87 MB, 1184x800)
3.87 MB
3.87 MB MP4
>>31049015
I got very addicted to image gen in 2022 when it started kicking off, but you do get over it. It's still good though, don't get me wrong.
>>
>>31049671
*shouldnt
>>
>>31049626

5090 is still top tier card for speed in AI gen as long as the model fits. If you are getting a rig just for AI, get one with a 5090.
>>
>>31049104


Yes, keep posting Black women. Great stuff.
>>
>>31049673

Valid point, image gen re-roll is more entertaining than video gen.

Could do that shit for hours. Such cheap entertainment :D
>>
File: uh_oh.webm (3.94 MB, 512x768)
3.94 MB
3.94 MB WEBM
>>31049671
See what you describe is the opposite of the experience anons have been sharing here for the past week. MiniMax takes direction extremely well and like extremely long, verbose, specific prompts and careful guidance. Autism, lots of autism. And you're totally right, too. I haven't seen this behavior before. I barely tweaked this prompt, ran it 4 times, two of them he does shit like this out of the blue. I haven't experienced this before and I barely touch anime stuff, I'm convinced you must be right.
>>
>>31049671
I can't get it to not include gibberish either. It might just be broken
https://huggingface.co/MiniMaxAI/MiniMax-H3/discussions/64
>>
>>31049697
I wonder if theres any difference if i put audio 1 nodes to audio 2
>>
File: 1786765750310548.mp4 (2.63 MB, 672x864)
2.63 MB
2.63 MB MP4
>>31048790
>Imagine not wanting some lights on during sex because the woman is so beautiful.


Yeah, because when it's 3am in the middle of the night and you wake up in the dark with a boner and you want to lift your girls leg up for that middle of the night quickie, turning the bright ass lights on, causing both of your eyes to uncomfortably squinch, because your eyes are adjusted to the dark, just so you can see your girls beautiful face is a must.

VIRGIN DETECTED.
>>
>>31045906
Holy shit this is hot as fuck, please post more hyper cock painful screaming shit
>>
>>31049730

Here ya go brotha. Buncha gens from the archive. Haven't updated it yet in last few weeks but buncha stuff in there.

https://drive.google.com/drive/folders/1tqhnrgfeM4XmX5chBIpiIO-CLFZrnYMv?usp=sharing
>>
File: Test 00026-Audio(5).mp4 (3.05 MB, 1376x768)
3.05 MB
3.05 MB MP4
>>31049601
Is that the full prompt? Part of the issue might be that you've just mixed FL2VA prompting with Ref2VA prompting then. You didn't even establish the subjects.

https://huggingface.co/MiniMaxAI/MiniMax-H3/blob/main/docs/VIDEO_PROMPT_WRITING_GUIDE_ref_en.md

Prompt for this gen in which I only went for the voice timbre:
https://pastebin.com/N4TZb0Pa
>>
>>31049811
That's what I get for rushing, I spent all yesterday doing ref2va shit and didn't think about it.
>>
>>31049811
Does it actually use the voice sample you inputted? Like if you changed it to a guy voice or miku voice, the actual voice would come to match that?
>>
>>31049626
nah, no way. at ~5k range you should absolutely aim for a 5090 (or if you can find a used 6000 Ada series with 48gb of vram)
>>
>>31047384
Fuck tattooed whores
They are a scourge
>>
File: ss6.webm (3.87 MB, 800x1200)
3.87 MB
3.87 MB WEBM
>>31049829
This is Carl Wheezer
>>
File: MiniMax_H3_00269_.mp4 (2.17 MB, 1056x608)
2.17 MB
2.17 MB MP4
I enjoy making japanese game show kino with this
>>
File: 1786740847577020.mp4 (2.99 MB, 992x736)
2.99 MB
2.99 MB MP4
>>
File: 1786734469188763.mp4 (3.53 MB, 1984x1088)
3.53 MB
3.53 MB MP4
>>
>>31049870
I've been loving all the japanese gameshow gens.
>>
>>31049498
Is it really that hard to accept that you're doing something wrong?
>>
>>31049873
tasty looking nipples
>>
>>31047384
So did you just use the video as input ref and prompt "regenerate same video without tattoos" ??
>>
>>31049743
kek good stuff anon
>>
>>31046851
Damn this is the best anal one I've seen
>>
File: 1000004687.mp4 (1.07 MB, 464x688)
1.07 MB
1.07 MB MP4
>>
File: 1786510174353975.mp4 (642 KB, 480x720)
642 KB
642 KB MP4
>>
>>31044058
>>31045395
It really is elite
>>
File: Test_00036-audio.mp4 (3.89 MB, 1056x608)
3.89 MB
3.89 MB MP4
>>31049829
Yes, but it's not a drop in replacement. If you want to use a male voice, you'll have to specify that for example, or it'll somewhat adjust the timbre to fit the gender and language.

Enjoy her talking with Chapelles voice.
>>
>>31049958
r2v has a dedicated "video editing" mode. it took a few tries and I did have to get specific with the location of the tattoos.
>>
>>31049709
Ok, I'll grant you, in this one circumstance of nighttime desperation where you're both half asleep, it doesn't matter how she looks.
>>
File: 697697489440557.mp4 (3.74 MB, 544x960)
3.74 MB
3.74 MB MP4
>>
New thread: >>31050095
>>
>>31046631
you sound like a feminist cunt bitching about lara croft
>>
>>31049811
Better with the subtitles. Hate not understanding what they're saying. Sounds good though, but so did the other version.
>>
File: MiniMax_H3__00100_.mp4 (489 KB, 736x576)
489 KB
489 KB MP4
I can't get it not to add gibberish after the word "test", maybe it works for you because you have more complex scene and there are other things to do


subject_definitions:
<Subject 1> woman from <Picture 1>
<Audio 1> is the voice-timbre reference for <Subject 1> (S1).

summary:
[reference generation + audio reference] The target video shows <Subject 1> doing a pose and speaking using <Audio 1> as voice-timbre reference

retention_analysis:
<Subject 1> (appears throughout): fully_preserved - the woman’s appearance from <Picture 1> is retained.
<Audio 1>: reference - the target uses the voice timbre from <Audio 1> for <Subject 1> (S1)’s spoken dialogue.

detailed_description:
The target video is a static camera shot.
[Shot 1] At 00:00.000, <Subject 1> (S1) speaks: <d>[English] test.</d> and starts posing hands on her sides

overall_soundscape:
empty photoshoot room

non_diegetic_music:
N/A
>>
>>31050251
The summary probably implies she's speaking for the whole duration.
>>
File: MiniMax_H3__00108_.mp4 (215 KB, 544x384)
215 KB
215 KB MP4
>>31050276
you get weird overflow type bugs if you syntax it wrong in different ways, still don't know how to do it right
>>
>>31047161
can you provide an example of the prompts you are using to create the harry potter vids?
>>
>>31043970
seedance?
>>
File: giwtwm.webm (3.78 MB, 1088x1920)
3.78 MB
3.78 MB WEBM
>>
>>31046851
bro WHAT!?
>>
>>31048084
more anon
>>
more eva



[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.