>>6259399
need more of this, give more!!!

File: supes and dan turpin.mp4 (3.76 MB, 640x360)

3.76 MB MP4

>>6259679

>>6259352

>>6259352
I don't know if Trump in the oval office is analogous to anyone in the Power Rangers

>>6259489
Trump in the oval office is more like the megazord frankly

>>6259489
fiction and of course AI can align anyone with anything. It can be satire which obviously that is. It could make NetINyahoo as god and Trump as Jesus and have a dig at religion and politics if done in a clever way. One thing you have to remember, you don't have to take it too seriously unless it is packaged as such.

File: part-1.mp4 (3.68 MB, 768x768)

3.68 MB MP4

>>6259755

https://d.uguu.se/hHnYrXJR.webm

1/3

File: part-2.mp4 (3.66 MB, 768x768)

3.66 MB MP4

>>6259755

https://d.uguu.se/hHnYrXJR.webm
2/3

File: part-3.mp4 (3.64 MB, 768x768)

3.64 MB MP4

>>6259608 >>6259755

https://d.uguu.se/hHnYrXJR.webm
3/3

>>6259574
It said "fork up their ass" perfectly

>>6259614
would that not be legal?

>>6259623
I genuinely have no idea lol

File: Centaurs_falling_from_hat(...).mp4 (1.42 MB, 640x360)

1.42 MB MP4

All of my attempts at this prompt had issues with visual continuity, but at least this one had the timing down

>>6259676
sam altman is trying to bring sora back

>>6259442
this is neat. how was it done?
funny how much acromegaly there is though.

>>6259677
i hope it happens but i am not sure microsoft will offer a free version like before :(

File: 1b8777b6-6e55-47c2-a768-2(...).mp4 (5.73 MB, 864x480)

5.73 MB MP4

>>6259749 >>6259756 >>6259878 >>6260520

1/?

iterating on this shit for weeks and its still ass. perhaps there is no hope left for this project

>>6259690

That's OK. Just keep going.

>>6259572
>>6259573
>>6259574
Pretty good

>>6259690
Why are these so slow and ass?

>>6259690
If you are using an enhanced prompt and depending on what LLM you use for local (I assume you are trying automatic generation for the project) then it comes down to the base prompt, what instructions you want the LLM to use to generate the script and the dialogue etc. I also try to be able to generate automatically and using an enhanced local LLM. Really LLM's should be ablet to write scripts which are catered for the video gen model being used. Really the best way is to use a commercial LLM to write the scripts of which they are for the video gen length possible but each following script follows on from the previous so that essentially you are doing a continuous story line. Is it better to simply get a LLM to write a script for a story and then tell the LLM to convert that to say 20-30 sec segments in say the H3 prompt format? Are you trying to procedurally generate a script? Wit hsay Minimax H3 you do have to really use the H3 prompt guidelines, it will gen using a basic prompt tho, but as soon as you get away from just mostly static simple gens it just ends up a mess. You need the prompt to be following those guidelines, when people talk they need to know what they are talking about and if via a LLM and that LLM needs to know what the subject is.

File: 1b8777b6-6e55-47c2-a768-2(...).mp4 (5.82 MB, 864x480)

5.82 MB MP4

>>6259881

act 1 of 4

File: 1b8777b6-6e55-47c2-a768-2(...).mp4 (5.92 MB, 864x480)

5.92 MB MP4

>>6259884

>>6259879
act 2 of 4

File: 1b8777b6-6e55-47c2-a768-2(...).mp4 (5.9 MB, 864x480)

5.9 MB MP4

>>6259885

>>6259881
act 3 of 4

File: 1b8777b6-6e55-47c2-a768-2(...).mp4 (5.82 MB, 864x480)

5.82 MB MP4

>>6259884
act 4 of 4

>>6259878
>get a LLM to write a script for a story and then tell the LLM to convert that to say 20-30 sec segments in say the H3 prompt format?
this is a simplified version of my approach.
1. it generates an episode pitch
2. then breaks out the acts structure with story beats
3. then it molds that structure into storyboards to plot out narrative trajectory, evidence progression, dialog intention
4. then it translates those storyboards into production timelines, framing and timing shots, establishing specific actions and stage notes, etc
5. those production timelines are finally then packaged into chains of 15s h3 clip prompts

>Wit hsay Minimax H3 you do have to really use the H3 prompt guidelines
yes, I've used the official prompting guides along with a number of supplementary resources to produce the prompt generation guidance

>when people talk they need to know what they are talking about
the dialog is a mess through the script generation funnel, not from the final prompt itself. the LLM struggles to understand what is important for people to say vs what is important as visual language vs what is unknown to the viewer. I've iterated on the dialog generation a number of times but have only made small gains. it's been hard to figure out where to make meaningful improvements in this gigantic rube goldberg machine

>>6259905
there is no doubt about it whilst an LLM can write the script, can format it to H3 and simply use additional guidelines (you can instruct an LLM to do anything) the LLM and H3 video genning will break unless you do it a certain way. Unfortunately H3 and even commercial ones can't really go beyond the short genning sequences, you have to treat each genning sequence as something that will go into a cut-scene. You get continuity errors, as soon as the camera switches and then you go back some things might change, the more complicated the scene unless it's a close up of say a person the more likely something will change unless you use an image that can match what was in that previous scene. With regard to storylines again a script - the storyboard matching the dialogue has to be all created and has to make sense. This all has to be generated via say a LLM or written by hand just like they do with movie scripts. When people are doing longer AI gens they are clearly going to be doing multiple gens and also doing editing because mistakes happen and need fixing. Yes AI models will generate say a 30 sec sequence, how can the next sequence align with that one without errors unless you first view what was genned and maybe tweak the prompt for the next gen?

File: dragon ball zeta.mp4 (5.55 MB, 854x480)

5.55 MB MP4

>>6259924

>>6259679
Don't know. I saw it on /tv/ yesterday.
I think he uses Runway (Gen-3) or other models

>>6259919
nta, I think for Minimax H3 the face and body swapping could also do that. However of course the prompt would also have to be that it is all real life instead of it be real life character but still in the cartoon environment. There might also be other Loras that can do cartoon to real life and vice versa. The dialogue is simply lip synching so not a problem.

>>6259352
better than 8bitguy's movie

File: Trump.webm (2.9 MB, 640x352)

2.9 MB WEBM

Trump

File: Leonarda Jonie - This is (...).mp4 (2.89 MB, 480x270)

2.89 MB MP4

Trumpey ...

File: knightfall.mp4 (4.92 MB, 854x480)

4.92 MB MP4

>>6259924
Thanks for the explanation.

>>6260056
Yameroooo! this looks good.

>>6259677
If it is back, it is going to have boring censorship within a month like last time.

>>6259690
Don't give up, anon.

>>6260417
Cant this model do blushing?

>>6260454
Fuck now I want Axolotl waifu.

>>6259677
I will only care if it's $15 a month USD for 15+ generations daily, ZERO censorship, 15 seconds minimum each video, loli allowed.

File: Fast Food - 000001_6MB.mp4 (5.09 MB, 576x324)

5.09 MB MP4

>>6261121 >>6261125

Making random AI shit in Kling 4.0, which was just released.

It's okay, it gets really blurry during high-intensity action. Wan 3.0 is the best I've seen for action. Seedance 2.5 is also good, but it tends to have a lot of IP censorship so you can't have as much fun with it.

>>6261099
Fucking insane pricing, though. For the $10 "General Trial Package" (limited to 4 purchases) you get 100 units. That's exactly 100 seconds of 720p footage. Then it's $14 per 100 seconds. 1080p costs 50% more.

For $700, you can get 5,000 seconds, which is 166 of their 30 second videos!

That pricing is egregiously insane lmao

>>6261121
>>6261099
Oh, wait, I lied. Generalist Trial doesn't work for 4.0, only their Flagship Trial does. Whjich is $9 for 64 units. Which is...two 30 second videos.

$10 for 60 seconds of video. lmao.

>>6261125
>$10 for 60 seconds of video
at $100 for a 10min video, which qualifies for lfv rpm on yt, you only need ~10k-20k views for a video to pay for itself

>>6261132
Kek that's retarded thinking buddy.

>>6261140
how so?

>>6261142
Because with how harsh the censorship for AI content is on websites, you're not going to be making anything unless you have a pre-established channel, and that cuts out 99% of people. Plus you're simply goalpost shifting from the "Insanely expensive price for an alright but not the best" video generation by saying "bUt YoU cOuLd MakE mOneY!!11".

File: Snake and Akuma - 000003_6MB.mp4 (5.05 MB, 1328x996)

5.05 MB MP4

Older Kling 3.0 Akuma vs. Snake

File: Batman - 000009 - 1of2_6MB.mp4 (5.55 MB, 1920x1080)

5.55 MB MP4

Wan 3.0 Batman vs. Wesker, split for size

File: DBZ - 000006 - Vegeta pow(...).webm (5.22 MB, 1344x756)

5.22 MB WEBM

>>6261171

Applying 'realification' with Wan 3,0 to Vegeta powering up against Goku.

Not perfect, but it's fun.

>>6261167
didnt realify them in the slightest though. just added more aggressive shadowing

File: DBZ - 000007 - Vegeta pow(...).mp4 (4.88 MB, 960x540)

4.88 MB MP4

>>6261171
I guess you're right, it's more 3D-ify. I wanted it to move closer to real, but not live-action.

Here I tried to be creative and tell Wan to make Vegeta and Goku Crystal gods. Some of the really crazy prompts for 'style transfer' actually don't work well. I've been playing with this stuff the past week.

>>6259676
Kek. maybe AI isnt that bad after all

>>6261265
>>6261282
I've really enjoyed this concept, though I wish you were funny

>>6261163
welp, umbrella's fucked

File: 2026-10-30-02H22m56s Seed(...).mp4 (5.61 MB, 864x640)

5.61 MB MP4

>>6261321

>>6261287
I'm not hand writing the jokes simply giving an overall theme for the LLM (local 27bn) to write jokes and the conversation as well as the scenes. I do that to randomise things rather than having the speech already laid out for it to simply follow. Of course having the base prompt to instruct the LLM could be better, but also comedy is subjective and some would say Seinfeld was not funny anyway or at least hit or miss.

>>6261158
"Sir... there's only 450TB unique futa videos in the archives..."

>>6261303
nta but some of them do a reasonable job of mimicking the tone of seinfeld, just missing some zing. anyways that anon is just empty inside.

How did /aicg/ turn into capeshit & furries general?

>>6261603
the matrix guy's spirit broke and all that's left now is the sloppers

>>6261617
I might post today, but my spirit about to break to, I'm getting dejavu when I tried making something with midjourney a year ago, the lack of control is infuriating fixes one thing in the shot breaks something else like a monkey's paw lmao

File: WKquTIdjiAGHXH21.mp4 (1.34 MB, 480x852)

1.34 MB MP4

>>6261688

I've started noticing all these AI girl accounts popping up on social media.

They are getting better

File: q1HFn699Q8Y4L2oD.mp4 (994 KB, 720x1280)

994 KB MP4

>>6261686

>>6261603
Instead of complaining, why don't you post something better?

>>6261603
How did {this thread I am not contributing to} manage to not have what I subjectively like?!?!?!?!?!?

>>6261603
it's really boring innit. i miss the days of funny ai vids. i'm out

>>6262096
No you're not. You're 100% reading this reply right now.

File: h3_muse_draft_960x536_4chan.webm (5.07 MB, 960x536)

5.07 MB WEBM

>>6262140

draft, not sure im finishing this, cut from 4 failed attempts around 30s missing of stuff, did not fix audio also

>>6262104

What is it supposed to be?

File: H3 Hybrid B45 - Cafe reac(...).mp4 (1.61 MB, 1344x768)

1.61 MB MP4

>>6262140
An attempt to see if you can create a spatially coherent animation from a single image and references, the ultimate goal would be to just give it a script and have something on autopilot

>>6262025
kek