Stable Diffusion 3 API Released.

[Aitrepreneur]

A few precision:

This is NOT the "real" SD3 model, the API one is a much older model that is not gonna be representative of the final model because...well...the SD3 model is STILL in training and will be released when it's ready. The API one was probably released because of a contract with Fireworks AI who made the workflow for that version of the model.

So YES, for those asking the SD3 model will be free and open-source, it will be much better than what you see here and it will be released to the public when it's ready, so be patient yall.

[malch]

The music is very distracting, maybe have the music volume lower next time.

[obscuremusictabs]

Please no music. It sounds like another tab is open.

[scotadam]

The problem is not the music but the fact that there are lyrics.

[xd]

Gunna be hard to get me off 1.5 with my 50 extensions, 100 trained models, and 3000 Loras.

[OmriSadeh]

Was hoping you would show mainly images you created on sd3, especially if you’ve had prior access

[viddarkking]

I am honestly salivating for the next few months.
Once the Community has had the time to Fine Tune SD3 And Develop the best practices to train the New Models and Loras.
Things are about to get really fun.

[hakuhyo]

ELLA did such a great job in prompt comprehension to the point that it’s difficult to see what SD3.0 is adding, if quality of example is what to go by.

[Panda-ikuk]

I have been enjoying SD2 w/a1111. Will something like that every be created for SD3 so i can run locally, for free, as much as I want?

[mr_pip_]

In fact, apart from a further advance announcement, of which there have already been several, there is still nothing.
I'm curious to see when the models will finally come out for download so that you can really see what you can do with them. Until then, I find other developments more exciting at the moment.

[gdizzzl]

I just wanna tell everybody in the comments that we have enough anime porn to last us a lifetime so if you guys wanna focus on some other type of art, that’d be great

[supercurioTube]

I couldn't watch the whole video because of the volume to the background music with lyrics. Too fatiguing.

[RikkTheGaijin]

Porn. That's the main difference. SD can do Porn. The other closed source models cannot.

[AI_EmeraldApple]

I don't like it that SD3 currently looks bad compared to lykon's examples on his twitter page. I think it was a bad move to release a half-baked workflow version of SD3 that doesn't meet the aesthetic quality of MJ6. Looks like i'll be sticking with SD1.5 models for a while longer

[nicolas.c]

haha great info, and the joke in the middle made my day!👏

[TheBurningBuffalo]

In the sofa picture one of the dots disappears from 2:00 to 5:15. I wonder how good the text really is, how often they tweaked the pictures before releasing them.

[arothmanmusic]

Now that we appear to have a functional text generation, I'm curious about the implications of copyright for the fonts in the training data. AI companies are already being sued by creators of text and images used in the training data… are foundries the next to jump into the fray?

[silentrobcanada]

I'm a little worried about Stability AI's future with the round of layoffs and CEO departure. I hope whomever acquires them continues to keep the open source ethos.

[AIFuzz]

I think text implementation will be better overall. The initial base model will always be the “start” and people will often overreact at the quality. As time goes on and with improvements and fine tuning, the forthcoming forms of SD3 will be better.

[titankronos]

What's the point of sd3 if it as censored as mid journey and dalle 3, atleast Mj and dalle 3 has better image quality than sd3, i hope that a less censored version of sd3 will be available in the future.