My foray into AI

animagic · February 9, 2026, 4:32am

Time for me to contribute again to my own thread!

One issue that comes up is to maintain coherence with longer videos. One segment is between 10 or 15 seconds, so we need a way to make two segments fit together. One way to do so is to generate the video segments with start and end frames. The end frame of one segment then serves as start frame for the next segment. You could simply use the last frame from the generated video for that but I found you often lose image quality that way.

True to the purpose of this thread the staRT/end frames were created from iClone renders. I then used Nana Banana Pro to make that render more interesting and give it the look I wanted. However, it is difficult to control the AI with just a prompt to get consistency between start and end frames. These frames need to have some overlap in order for the AI to have an idea where the action takes place.

What I found working is to select a specific frame to serve as style. This style is the transferred to the specific image which then has a similar look as to appearance, lighting etc. I did a lot of experimenting to get to this point, but I think it was worth it. Sometimes the AI-generated images need post-processing, which I do in Photoshop. I will explain that in more detail at some point. The video segments were generated with Kling 2.5. Sound was created with Krotos Studio Pro

Anyway, for now here is the video:

animagic · February 11, 2026, 9:53pm

I’ve been working on my taxes all morning, so it’s time for a creative break. Very recently Kling 3.0 was released and became available on Freepik. They now also provide Grok for images. I wanted to compare Grok with Nano Banano Pro and used my usual prompt an a photo of a young woman.

The prompt is: “Change semi realistic 3D cyberpunk style”, and here is the result:
Nano Banana Pro:

Grok:

So Grok is a bit more subdued.

I used Kling 3.0 to create a clip of the young woman drinking from the glass. Then, just for fun I decided to used the Grok render as start frame and the NBP render as end frame, and then do the reverse. It resulted in an interesting morphing:

AutoDidact · February 12, 2026, 12:53am

Nice!!
very cool effects
This video is a bit of a chilling account about how fast the AI models , in general, are progressing.

Sophus · February 12, 2026, 11:44am

In those videos or reports I only see people AI-generating all the same looking Web apps each time, which don’t do anything new. I don’t get it. Nobody needs these apps. It’s feels like when people, who used Microsoft Frontpage for the first time 25 years ago, said, that web developers will be replaced by it.

AutoDidact · February 12, 2026, 12:58pm

in those videos or reports I only see people AI-generating all the same looking Web apps each time, which don’t do anything new. I don’t get it. Nobody needs these apps.

Alot of people are "vibe coding "custom productivity apps that deal directly with their individual business based on their individual information databases and existing websites and other internal media.

They’re using AI to increase their personal productivity in a highly customized way.
And they’re doing so with little to zero coding knowledge and not having to go out and hire an app developer
or buy an existing commercial office productivity software from Microsoft or the Apple apps store or similar.
That’s the advantage here.

It’s feels like when people, who used Microsoft Frontpage for the first time 25 years ago, said, that web developers will be replaced by it.

False comparison.^^
Building a website back in the days of Microsoft front page still required you to have a fundamental working knowledge of HTML another web development code if you need something truly useful with the interactivity of an e commerce site attached to a Data base

This is a far cry from what we have today as you can literally sit and tell the AI
( in plain laypersons english.)
What you would like to have built into your customized app and have it create all of the graphics & widgets, controls and arrange them in a way that’s going to be deployable across a wide variety of screens & devices.

Remember way back in the days of Iclone 7, when Reallusion announced the addition of a python API?

Notice how we did not see a huge flood of interesting and affordable third party python plugins to add functionality to iclone that many people were asking for.
And why was this ?

because people in this community had to actually have expert knowledge of python code to be able to create a plugin.

The video below is very long but the link cuts in at 18 minutes and you’ll see exactly why people are excited about being able to build their own software tools simply by describing them to an AI like Claude from Anthropic.

planetstardragon · February 12, 2026, 4:12pm

i believe that what we are witnessing is an information science phenomena called
Data Gravity - the more mass of “intelligence and data” that accumulates, the greater the gravity. the more you know, the more you can understand, to learn more - rinse wash, repeat. and how are blackholes formed ? lol

What is data gravity?

Imagine a world where data acts like a planet, exerting a gravitational pull. This is the essence of data gravity — a term that captures how large volumes of data attract applications, services, and even more data. As the pool of data grows, so does the strength of its pull, deeply influencing how and where subsequent data will accumulate and interact.

Sophus · February 12, 2026, 5:42pm

He simply generated a page that shows YouTube videos. I don’t see a business here. The business is most likely to sell vibe coding courses.

Not really, back in the day SQL or Visual Basic were marketed similarly, where the business owner “didn’t need” any database or app development knowledge because you can simply write a query or drag and drop a software together, anyone could do it.

Clearly, most of the mentioned statements in the video were only made to hype up investors because it’s all a house of cards.

K_Digital · February 12, 2026, 5:44pm

@Sophus

After much suffering I have managed to reduce by at least 1/3 the VRAM requirements of LTX 2, which means:
10s at 720p can be done with only 8GB of VRAM

I still havent found out how to … can you give us a brief overview on how to ?

AutoDidact · February 12, 2026, 7:15pm

I don’t see a business here, The business is most likely to sell vibe coding courses.

You can code a tool for any business.
it could be a business to sell any product like an expensive $200, NVIDIA dependent, “Accuface” plugin when I can do the same thing free FREE with blender.

Perhaps you do not have a personal use case for a new software dev tool but that does not mean its useless to every one else.

because it’s all a house of cards.

…and let me guess when it “crashes” every LLM ,AI system will be deleted from the planet and the entirety of human civilization will collectively pretend that this tech never existed
NVIDIA will sell us affordable GPUs again and we will all have go back to figure out how to use UE5 and still fail to make animations like this.

BTW animagic started this thread to discuss AI animation tools/ methods
You are clearly an Anti AI activist trying to hijack this discussion
Maybe consider starting your own thread on why you hate this technology

animagic · February 12, 2026, 8:45pm

Thank you. I keep a positive outlook and I appreciate contributions by others. There are many opportunities to join discussions with a different perspective.

I feel like an elder sometimes, seeing the the young-ones rolling over each other. One angry comment I read the other day was that the term “AI” is just a marketing ploy by Big Tech to lure the overly gullible into something. I had to comment that the term AI is from 1956 and introduced during a workshop at Dartmouth on the subject. Ever since computers came to the fore, the question has been raised whether they can think. Turing actually introduced a test to see if a subject could distinguish between a machine and a human. If you are a philosopher you might wonder “What is HUMAN intelligence?”

More than forty years ago, I took courses in AI as part of my curriculum. Every time when I see another proclaimed “god-father of AI” coming out of the woodwork warning us, I smile because he is so much younger than I am.
It helps to know history (real history) as it puts things into perspective. When I learned about AI, there was also a fierce debate between hard AI and soft AI. The question was if you could encode “knowledge of the world” by building a large rulebase, or perhaps by building semantic nets. But it had to be painstakingly hand-coded. On the other hand there was the idea of neural nets, which were intended to model the brain and learn by themselves.
It is a very interesting area of research I think. I worked on machine translation and even though you can do a syntactic analysis of a sentence, it is not sufficient. There may be more than one analysis, and words in the sentence may have multible meanings. We didn’t have LLMs, but the idea started to float to include statistical methods: in this context it is more probably for a word to mean A. Even though it is fun to write translation rules (if you like coding) we started to include a knowledge bank of sentence fragments and their translation. Other groups worked on similar approaches. We needed several Sun workstations to run our stuff; today all that would be much easier.

Today, we have Google Translate of course, but behind it is a similar idea.

Anyway, I’m glad I am retired and don’t have to worry about my job, which is a concern I’m sure, but automation always has been.

AutoDidact · February 12, 2026, 9:47pm

Anyway, I’m glad I am retired and don’t have to worry about my job,

You and me both my friend frankly I made it out just in time it seems (Dec 2025)
Back during my former career as a graphic designer for print part of my job was to make sure a file such as a magazine layout was properly formatted to be sent to the printer which was a fairly detailed process.

I left that field for 3D animation in 2008

Last year I was investigating the possibility of getting some actual hard copy prints of my graphic novel that I’m selling for Kindle on Amazon.com.

I admit I was a bit taken a back by how automated this print pre flight process had become with all of the online print on demand services
All of whom required that you simply upload all of the pages of your graphic novel as high quality tiff files with no pagination or other layout formatting.
One service even turned my single pages and into complete virtual book whereby I could flip through the pages with my mouse cursor right there in their free preview utility.

I personally am not one given to making “New Year’s resolutions”, but I had sort of committed to making 2026 to be the year that I stopped arguing with anti AI people on the internet.
but sometimes it’s hard not to get drawn in.
I mean, understand I am not some sort of a utopian who believes that AI is going to bring us some sort of Star Trek future.
And yes, I’m in a position where I never have to worry about the state of the job market ever again, particularly in any sort of creative field.
But it is hard not to resist responding to people who jump into every AI, discussion and basically assert that any of us who even dabble with the technology are somehow, foolish, wasting our time, it has ZERO use case value or we are even evil people because this technology is on the verge of collapse and is going to completely disappear from human civilization any day.

Sophus · February 13, 2026, 10:26am

No, I suggested Wan2GP a couple of posts above. I just don’t fall for this “Silicon Valley Apocalypse” talking points that don’t come from any software developer folks but only from amateur vibe coders and out-of-touch CEOs and marketing people. Don’t believe the hype. That’s all what I’m saying.

StyleMarshal · February 13, 2026, 2:54pm

AutoDidact · February 13, 2026, 6:14pm

The six finger trope would have been mildly funny…….three years ago
We are in a different world today ,so the people praying for the American AI companies to fail should know that China is not standing still so the tech will continue to develop

Seed dance 2.0 from the makers of tik tok

StyleMarshal · February 13, 2026, 6:29pm

yeah , Seedance 2 and Kling 3 are the best AI generators right now.
Next updates of Sora and Veo will follow…

Crazy what happens , the development is so fast and strong.

AutoDidact · February 14, 2026, 12:35pm

More seed dance.
Hollywood production & traditional 2D/3D animation & VFX will get alot cheaper to produce

Sophus · February 14, 2026, 4:57pm

Yes. I assume you haven’t installed in yet, correct? Here is a step-by-step installation. Not sure if you want to do this, but might be worth a try. Make sure you have enough SSD storage left (200 GB). There seems to be some problems with Out-Of-Memory-Errors in newer version of Wan2GP since January. Not sure if this was already solved.

The easiest way to install it is to go to https://pinokio.co/ and download the Pionkio app.

Put the exe into a to folder of your choice. I highly recommend a directory without any spaces in the name, because this can break some apps or scripts from working correctly. For example, C:\Pinokio\ is fine.

Next, start Pinokio.exe. This will take a minute as files will be extracted into the folder. After a while the Welcome Screen should come up.

Next, while Pinokio is running, go to https://pinokio.co/ click on “Wan2GP” and one the next page on “One-Click Install with Pinokio”.

Allow the browser to open Pinokio. You’ll then see a list of required software you need to install.

Click on the Install button. This will now take a pretty long time to download and install all requirements.

After a while a pop-up will ask you how you want to name the Wan2GP installation. You can leave it as wan.git and click on Download.

On the Install screen click on Customize.

On the Customize screen there are three temporary paths for downloads from Gradio (browser user interface), Huggingface (online storage space for AI models) and Torch (deep learning library). Pinkio defaults them to ./cache/ inside the /api/wan.git project folder. If you want to save space and share files with other projects you can delete these lines. This will move the cache folder to the main Pinokio folder. If you want to try out other projects which use the same files and don’t have a lot of SSD storage, this is recommended. Just the base installation of Pinokio and Wan2GP will take up about 35 GBs, without any video models downloaded, which come on top.

Next, click on the Install button at the top.

Next, the project window for Wan2GP inside Pinokio will open. If still some requirements are missing, for example CUDA, click on Install. Again, these might be multiple GBs of data and can take a while.

After the requirements are installed you see the message, that Pinokio was disconnected. You should see a start button at the top of the console while Install is highlighted in the menu on the left side. This means you can start the install of Wan2GP. Click on the Start button to start the installation of Wan2GP. Again, this will take a while.

After the installation has finished you see this screen. Click on Done.

Click on Start in the left menu, to run Wan2GP. The start-up will take some time, it might need to download more stuff. After that, you’ll see the Wan2GP browser interface inside Pinkio.

You can switch to a web browser by clicking on the small icon right next to Open Web UI and only watch the terminal in Pinokio to see what happens in the background or to see error messages.

Inside Wan2GP you can change the RAM/VRAM profiles under the configurations. This is very important, depending on your computers hardware specifications. But like I mentioned above, some users reported issue with OOM-errors if they don’t use profile 1 in the configuration, which defeats the purpose of Wan2GP since the other profiles allow for much lower RAM and VRAM requirements.

On the main site you can now select an LTX-2 model and simply try it out with the example prompt. The model will be downloaded if you run the first generation. This will again download dozens of GBs of model files.

The good thing about Pinokio is, that everything is contained inside its folder. If you want to delete Wan2GP, simply delete the wan.git folder inside the api subdirectory. If you want to delete Pinokio, simply delete the Pinokio folder.

StyleMarshal · February 15, 2026, 7:40pm

Crazy ! Bounty Hunters?

AutoDidact · February 15, 2026, 10:00pm

I am probably one of the biggest AI. enthusiasts here in this entire community.

But the narrative being spun by the maker of the above video is a false one.

My biggest gripe, with my fellow AI, enthusiasts, is their seemingly unbreakable addiction to other peoples, intellectual property and trademarked characters.
What is the point of saying that AI is going “to liberate us from the big media giants” when you don’t have the imagination to sit down and come up with your own original characters and stories and use AI to illustrate and animate them.

I created my own private Facebook group specifically for original comics and graphic novels created with AI. And the guys over there are coming up with some great original heroes and stories.

Now yes, I admit it.
I have some AI generated content based on the HALO video game universe on my youtube channel.
But I am just a nobody Fanboy with 1500 subscribers and cannot even qualify for monetization on YouTube and also the HALO video game franchise is pretty much a dead property at this point.
So nobody at Microsoft is sending lawyers after me and my little fan boy videos
Getting less than a 100 views per video.

However, my original graphic novel, that I am selling on Amazon ,is a completely original character in a completely original universe that I’ve developed and any future graphic novels and 2D animations that I make with those characters will belong to me.

Also the narrative of this video is claiming that big companies in Hollywood want to “destroy this technology”

I think the exact opposite is true.
In fact, you need look no further than the fact that Disney once sued stability AI over unauthorized use of some Star Wars characters, but then turned around and invested one billion in open AI, giving them permission to use Disney characters for their sora app so that Disney can collect a royalty on those usages as well as crowd source the cream of the crop from those sora AI fan made videos and monetize them on their streaming service, Disney +.

The same happened when Warner music sued the AI music service Suno AI .

All of the luddites squealed in celebration that suno was going to be destroyed and it turns out that Warner music just wanted to partner with them to make sure they would get some royalty fees to pay the artists whose music was used to train the data.

So to all of the frustrated fan boys who have been crying for years about how Hollywood is “ruining their franchises and destroying their heroes.”
Well, this is your chance to compete against them with your own heroes and stories and stop complaining and stop regurgitating IP that doesn’t belong to you.

animagic · February 16, 2026, 8:18pm

Hollywood has long wanted a monopoly on film-making, distribution, etc. and that has slowly been eroded. They don’t like that at all. And now AI allows people to make movies outside of their sphere of influence. But, I agree with Autodidact, you need to put some work in developing your own stories, featuring original characters and the world in which they operate.

Banning models is of course not a solution, as it becomes more and more viable to run models locally.