RAMTHRUST in the wild! Good to see ya :)

@colinw2292823 I am everywhere and nowhere. New Phone, who dis? :3

Hopefully not, if it’s only online and heavily censored only once using it will bethe artsy types and I don’t know that they constitute a large enough population to justify this strategy. Hopefully they will continue to do what they’re doing which is provide open source models that are so good no one can compete. Off-line is the only way I’d use Wan.

they work with crowdfunding so i doubt they changed their political strategy. The only way they change that is to be sold to an American GAFA company..

I mean i also have that same philosophy, but I didn't they state during the presentation that weights would be available eventually? Didnt they do this with Wan 2.2 as well?

Well, this is the general direction the AI is going, specially with models developed by American companies. But I don't think this is the case. They did the same with 2.2, first a release online to get feedback and fix issues, and a bit later they opened the weights.

This is what I'm always commenting on the contests CivitAI organises (especially video ones) if people use only Commercially Available tools it's not promoting the consensus of AI Tech but only of big Corpo money greed.

Well that is up to alibaba I guess xD there are rumors about an open source release, as far as I know, but we will see...

Now! :)

@theally unfortunately it doesn't work for me, it's keeps saying that it's unavailable, do i need a membership ??

@Asurasef You can't use it to make NSFW content, unfortunately. It's an API only model - and neither the creators nor Fal.ai (who are running it for us), allow NSFW content.

got it, thanks for the clarification, hope the creators can allow it in the near future

rip

sad part is even when things get supported people run with the money, pony v7 was supposed to be out around January and it's been silence since, then illustrious v3 was supposed to be released in the summer after reaching a dono goal then last minute they decide it would be exclusive to their on site generation.

The models also keep getting bigger, at some point it's going to be very hard to keep up without an expensive RTX 9090 and 128GB of VRAM xD

@Tetsuoo 64G VRAM should be enough for everyone, who has less than 256G VRAM? 512G VRAM or gtfo. 1024G VRAM minimum, 4096G preferred.

Which is crazy when you consider how they just scrape the whole internet to train their models.

It's time to DUMP NVIDIA!

Just give it time. They can hold out all they want, the internet has a way of forcing the issue whether they want it or not. Could you imagine paying money for the dogshit on Netflix?

It will be interesting to see how local AI looks in 2030!

@Codeman119 Same wan2.2 as now, but with more loras and merged models😂

The Chinese are hell bent on undermining the USA with AI and are totally open sourcing everything so you will have access. They made DeepSeek for a million dollars or something and model size growing to ridiculous levels is just typical American mentality of keep doing what you think works but only works for you and completely ignore the issues others have as those issues are not their issues because they have access. And make note of things you hear as you can learn how the systemic Narcissism works as it was and has been the American AI companies that stated at the beginning that we want to democratize AI. So what happen? Nothing they got you to play with it so they could use you to further train and further test and once they got where they needed to be and put restrictions on everything and gave you API numbers to use with your paid account and yes grew models to ridiculous sizes requiring hardware that is absurd prices which to be fair is also the ponzi scheme crypto currency peoples doing too. Where there is a will, there is a way and now we have quantized models and speed loras. If you havent looked into this Pusa lora concept it actually proved that size isnt always how you get better results and they turned the 2.1 T2v wan into being capable of doing everything you can do with any other model and it works. So control camera, control movement, control poses, does image and text to video and makes still images too that are as good as flux. I thought too good to be true and tried it and sure as shit it looks better than WAN 2.2 and used their own 2.1 to accomplish it without a high and low model. It's just not all that user friendly being on Diffsynth and the wrapper set up on comfy is only possible because of KiJai but is still ideal for the 24GB+ Vram people but is promising concepts they are proposing can be done to all models. So don't fear! The rise of the meek is upon us haha figured i had to go Biblical at the end.

I think so far we need to wait they said 2.5 is in preview so lets at least wait until it final to see what happens.

@dillion1920 All for just a 4 sec of video, the current state of gen AI is pathetic really

@kunde2 The final version is a big lie. They will never finish anything. The new wan, will be preview version again and again.

Maybe, but I'd say there's plenty of people like me who just ignore what isn't open source. Figured I'd check out Wan 2.5, and suddenly I don't care because it's online only and not even with the option to purchase with buzz...Eh, Off to something else. haha

Don’t be greedy (lol), WAN 2.2 is more than enough right now for anyone who wants to create ultra-realistic videos or stylish 3D NSFW content, lol. Check out my page to see what I’ve made with WAN 2.1, and even more with WAN 2.2 using only the low-VRAM FP8 high and low models I released recently. I call it WAN 2.2 ULTIMA FP8 14B T2V/I2V. :)

Unless there is a breakthrough in AI, no way the amount of computing required to train those models is sustainable.

that's what I'm wondering.

Never! They wanna go commercial now...

The lie has poisoned corporate minds again: they think they stopped AI before it truly became open source. And they'll stop making GPUs with more than 16GB of DDR6, possibly keeping most cards between 8-12GB. Getting a machine with unified memory will be prohibitively expensive. The newest models will all be locked behind walls similar to paywalls or pay-to-win games. But all that will do is slow the public down. Wan 2.2 is the final "free" model for Wan, but we've already got plenty of open-sourced models and software. All we really need to do is find something like the old Raspberry Pi hardware licensing, and rally around getting Qualcomm or someone else to make open-sourced AI boards, maybe focus on building-up CPU drivers again, and engineer boards with royalty-free licensing for production. Build it all around an ARM-like processor, which can work with LPDDR5X (or SOCAMM2). We need to fight to keep this genie out of their bottles.

@Lacerna_Nebulae Yup. Either we scavenge our way into it, or it's the literal end of the world with powerful AIs in the hands of a select 'responsible' few. It starts with the little things like this. Get good, y'all! Vive la Résistance!

@MrSmith2025 Who wants to go commercial? How did you figure?

@essence25 WAN 2.5 is not open source and useable for us on local PC... You need to pay for it. Thats called "commercial".

@MrSmith2025 I saw the other posts on Reddit, but it does not specifically state that it will "never" be released, it is a little vague. These keep evolving so 3.0 comes out they will release 2.5. Maybe we should not freak out and be grateful for the work they are doing.

@essence25 I don't care. Wan 2.2 is good enough for private use...

我认为等他们有更好的模型诞生比如wan3.0之类的他们就会开源上一个版本的模型

Sora v2 is way better and can be run locally but only supports linux

trvke

cough online only cough

@TheP3NGU1N huh?

@Jaunty its api only.. its going to block nsfw.

@TheP3NGU1N oh

Check LTX 2.3 out, it's way better than Wan 2.5.

@TheSiteCantHandleFeedback Why would we ban a dude for criticizing Alibaba?

He was banned for breach of Civitai's ToS, specifically harassment and anti semitic, abusive language.

thank you for proving my point, I screenshot it before you nuked the thread and banned me like I said :)
never cursed, didn't insult anyone, didn't harass or use abusive language.