Do you agree? SD3 model release. But hands are bad?

[FusionDeveloper]

I heard there will basically be zero custom SD3 models, due to strict licensing rules that prohibit it.

[Powerlevelover]

From what I heard, the commercial licence on sd3 is different than base sdxl, right ?

So I am not sure there would be that many fine tunes.

I just hope the Pony model based on SD3 drops (the Sdxl version is really amazing it can do great SFW images as well)

[AIKnowledgeGo]

The primary issue with SD3 is that Stability AI's licensing deters most creators from developing community models. Without changes from SAI, SD3's prospects are bleak. While Stable Diffusion 3 excels in areas like pixel art, 3D art, and general artwork, it falls short in producing photorealistic images.

[vitalis]

Why would any community member spend resources in developing a community model, not just because of the strict licensing, but simply because the biggest platform Civitai banned all SD3 models?
Also, this video is just weird, when he was the one who broke the story about SD3’s strict licensing.

[sneedtube]

Sure but then SD3 is just a leap in performance vs quality (for a base model) and nothing even remotely groundbreaking, basically disappointing.

[galzajc]

i hope people improve it. it's not just hands basically any human/animal/monsters, ... anatomy. buti heard emad said it must have happened durin censoring phase. so it's probably no that hard to make model have those qualities again

[headscout]

Yes. When SDXL first released, it also bad at drawing hands. That's why nodes like Hands MeshGraphomer and any other ControlNet exist.

[brootalbap]

But Olivio: Why should SD3 achieve what lightning and XL couldn't ? 1.5 still is the most used and most supported by finetunes and loras.

[ghostbusterz]

Does this not contradict what you said about SD3's license? Why would a model with a garbage license become the standard?

[rando]

No commercial license = dead in the water. Add to that the fact that it’s not much of an improvement over SDXL and why use it? We need to move on to an actually open source company that doesn’t censor its models and that has a focus on output quality and not gimmicks like refiner models that almost no one uses anyway.

[pastuh]

I plan to train the model, but I definitely won't be modifying the base model. I am waiting for at least one community version

[zmtnx]

This aged like fine milk... SD3-Medium is too large, perhaps this would be true if they released SD3-Small.

[stevencrawford]

High hopes is all you can have at this point. 🤷‍♂️ Mid journey and Dall-E are starting to walk away with things. I can still keep up, mostly, with highly customized SD 1.5, but you can see that its not going to hold up another year.

[cowlevelcrypto]

Yep, it fails with a lot of things in Comfy. I hope we get support before S-AI completely falls apart. It might be the "new standard", but i doubt many people will abide by the current terms. Stability has no mote just like Google.

[pirobotbeta]

I'm gonna stick with SD 1.5...it does what i want and I can play around with it without having to delete models or anything.
Yes, I am voting to stay in the past...

[TheDocPixel]

SAI will sold or bankrupt by Christmas. No fine tunes that people want will be made. The core framework is censored which will make it almost impossible to do better humans. At least the ones that people want to generate.

[Dj-Mccullough]

The only way Stable diffusion 3 will become better is through piracy. People will refine the model, but they wont deal with that licence agreement.

[BuzzJeux_Studio]

Hi, yes hands are really really bad in SD3!

[TheMartiansLoveKuato]

Base SD3 is better than base SDXL. Looking forward to the tech to get better. But it is laughable to monetize SD3 in its current form.

[peacefusion]

but sd3 is rellying on the community to improve thier models. they want to hurt open source, they can fix thier own hands.