Stable Diffusion, every mention
77 scenes, the whole family · ← back to Stable Diffusion
tap a year for its mentions
every year anyone comfyanonymous (Comfy) 29Shawn Wang 23Batuhan Taskaya 14Suhail Doshi 6Ben Firshman 5Alessio Fanelli 5RJ Honicky 3Jed Borovik 3Eugene Cheah 3Jeremy Howard 2
Verbatim, from the transcripts: the passages where Stable Diffusion comes up
Why AI Labs With Unlimited GPUs Still Fail — Anjney Midha, AMP
- ▶ 37:25 Anjney Midha Like, you, you are an athlete of the mind, and you perform at the highest levels, and to get there, whether you're, you know, Anastasius or Waylon at Berkeley, or you are Robin, who, with Black Forest and created Stable Diffusion, or if…
Inside xAI: Building Grok Imagine in 3 Months, Videogen vs World Models, and Video Agents— Ethan He
- ▶ 1:20:45 unnamed speaker If you've used stable diffusion one or whatever, if you've seen the prompts, like,
Captaining IMO Gold, Deep Think, On-Policy RL, Feeling the AGI in Singapore — Yi Tay
- ▶ 9:50 unnamed speaker Then all of a sudden, Stable Diffusion came out, and you can run it on your laptop, and that was my learning rate.
⚡ Inside Google Labs: Building The Gemini Coding Agent — Jed Borovik, Jules + AIE CODE Preview
- ▶ 2:47 Jed Borovik You know, when Stable Diffusion came out, that to me was the first, like, Gen AI mode. 3 times in the scene
A Technical History of Generative Media
- ▶ 4:51 unnamed speaker You know, you cannot, I think everybody knows stable diffusion, you know, and then you have maybe, like, the flux models, and then you have Black Forest Labs.
- ▶ 4:59 Batuhan Taskaya History-wise, I think the biggest, like, the initial hit was stable diffusion 1.5, which is when we actually pivoted into this new paradigm of fall, generating media cloud. 3 times in the scene
- ▶ 5:32 Batuhan Taskaya Uh, Stable Efficient 2.1 came, it was a bit of a flop, so it didn't like, you know, got that much attention.
- ▶ 5:37 Batuhan Taskaya And then SDXL came, which was like the first major model. 3 times in the scene
- ▶ 5:51 Batuhan Taskaya People started fine-tuning their faces, their objects, whatever, and generations with this, like, LoRa's started to become very popular, and then after Stable Deficient XL, there was, like, a bit of a quietness around it, you know,…
- ▶ 7:26 unnamed speaker You know, it's not a trivial decision, but obviously the right one at the time, I would say like a lot of people were hosting stable diffusion, right? 3 times in the scene
- ▶ 10:59 Batuhan Taskaya So it's like, we were at like the right time, you know, the very, it was like, uh, the, the, actually like the, the, the space was actually so, so much worse than what we have today, where like the running basic, like stable diffusion, 1.5…
- ▶ 12:44 Batuhan Taskaya You know, when we first started, there was a single model, Stable Diffusion 1.5. 2 times in the scene
- ▶ 13:04 Batuhan Taskaya The next thing, like, you know, with the, like, with adding more models, you know, like Stable Diffusion XL was a different architecture, Pixart was a different architecture, all these, like, different architectures started coming around.
- ▶ 15:04 unnamed speaker Like, if I just take stable diffusion, right, and I put it.
- ▶ 26:18 unnamed speaker I was going to show, I'm going to pull up the Quinn MMDIT where there's like this dual streaming thing, which I last, I think SD three had it. 2 times in the scene
- ▶ 30:36 Gorkem Yurtseven And then people caught up within months and then stable diffusion was even maybe better or just as good as the lead, like a couple of months later and it was open source.
- ▶ 39:26 unnamed speaker Stability did not make money from stable diffusion.
- ▶ 47:56 Batuhan Taskaya You know, like when you see these cool LORAs, like you, like there's, we have like still a lot of people using STXL with their own LORAs because they're
- ▶ 50:54 Batuhan Taskaya The thing that we saw is, as the models get better, the, like Confu UI was a much bigger thing or like, you know, relatively much bigger thing in two years ago, like a year ago, when the models were like, you know, one of the biggest Confu…
AI Video Is Eating The World — Olivia and Justine Moore, a16z
- ▶ 0:56 unnamed speaker Latent Space itself was started because of Stable Diffusion, and then there was, like, yeah, improvements in, in image generators for a while.
- ▶ 1:41 Justine Moore Generative media, too, with Stable Diffusion in, I think it was, like, September, 20, twenty-two-ish, and it's grown so quickly since then, and Olivia and I talk about this all the time, because it used to be, um, that our friends in…
The Agent Network — Dharmesh Shah, Agent.ai + CTO of HubSpot
- ▶ 1:33:59 Dharmesh Shah So he's kind of gone deep on, uh, stable diffusion and the algorithms and things like that. 2 times in the scene
AI Engineering for Art - with comfyanonymous
- ▶ 0:50 Shawn Wang I would say that when I first got started with Stable Diffusion, the star of the show was Automatic one-eleven, right? 2 times in the scene
- ▶ 7:46 comfyanonymous (Comfy) Like for example, you couldn't use like Excel and SD 1.5, because those have a different latent space, but 2 times in the scene
- ▶ 7:46 comfyanonymous (Comfy) Like for example, you couldn't use like Excel and SD 1.5, because those have a different latent space, but
- ▶ 8:06 comfyanonymous (Comfy) That's the problem that that's the, the reason why stable diffusion actually became like popular, like, cause was because of the latent space.
- ▶ 8:47 comfyanonymous (Comfy) So the, the reason I was hired is because they were doing, uh, SDXL at the time and they were basically, SDXL, I don't know if you remember, it was a base model and then a refiner model. 3 times in the scene
- ▶ 9:11 Shawn Wang But they didn't, they didn't pursue it for, like, SD three.
- ▶ 10:08 Alessio Fanelli So stable diffusion obviously is the most known. 4 times in the scene
- ▶ 10:17 comfyanonymous (Comfy) Well, the, the latest, uh, state of the art, at least, yeah, for images, there's, uh, yeah, there's Flux, uh, there's also SD 3.5. 4 times in the scene
- ▶ 11:31 Shawn Wang There's a lot of community discussion about the transition from SD 1.5 to SD two and then SD two to SD three. 3 times in the scene
- ▶ 11:31 Shawn Wang There's a lot of community discussion about the transition from SD 1.5 to SD two and then SD two to SD three. 2 times in the scene
- ▶ 11:31 Shawn Wang There's a lot of community discussion about the transition from SD 1.5 to SD two and then SD two to SD three. 6 times in the scene
- ▶ 17:08 Shawn Wang And then like, there was like other methods like textual inversion that was popular at the early SD stage. 2 times in the scene
- ▶ 19:29 comfyanonymous (Comfy) That we're trained for SD 1.5. 3 times in the scene
- ▶ 19:31 comfyanonymous (Comfy) They also kind of work on SDXL because SDXL has the, has two text encoders and one of them is the same as the, as the SD 1.5 clip L. 2 times in the scene
- ▶ 19:42 comfyanonymous (Comfy) So those, they actually would, they don't work as strongly because they're only applied to one of the text encoders, but, uh, and the same thing for SD three, three, SD three has three text encoders. 3 times in the scene
- ▶ 20:44 comfyanonymous (Comfy) So, but the hack that, uh, like you've, if you use stable diffusion 1.5, you've probably noticed, oh, it still works if I, if I use long prompts, prompts longer than 77 words. 2 times in the scene
- ▶ 42:04 comfyanonymous (Comfy) It's basically, what they did is they took SD two, and then they added some temporal attention to it, and then, uh, trained it on videos, you know.
- ▶ 43:55 comfyanonymous (Comfy) There's one that released at the same time as SD, 3.5, same day, which is why I don't remember the name. 2 times in the scene
- ▶ 46:14 comfyanonymous (Comfy) Then I made, basically, yeah, they hired me because they wanted the SDXL, so I got the SDXL working very well in ConfiUI because, yeah, they were experimenting with it. 4 times in the scene
- ▶ 49:44 comfyanonymous (Comfy) Like stable effusion models, like at least the, the open source side, and like, it's going to be best way to run or models locally, but we will have a few, like a few things to, to make money from it, like, uh, cloud inference or like that…
2024 Year in Review: The Big Scaling Debate, the Four Wars of AI, Top Themes and the Rise of Agents
- ▶ 48:07 Shawn Wang So there's no need for the stable diffusion or comfy UI workflow of like mask here and then like infill there and paint there and all that, all that stuff. 2 times in the scene
[Paper Club] Intro to Diffusion Models and OpenAI sCM: Simple, Stable, Scalable Consistency Models
- ▶ 1:53 RJ Honicky So it'd be useful to talk about what we're talking about here and, um, and in, in reference to, you know, sort of stable diffusion, um, and latent diffusion.
- ▶ 23:31 RJ Honicky I think that, uh, like the SDXL three and flux, but those are not this type of model.
- ▶ 23:31 RJ Honicky I think that, uh, like the SDXL three and flux, but those are not this type of model.
- ▶ 23:49 unnamed speaker Flux is, is great for text as well as, as the, uh, 3.5.
Answer.ai & AI Magic with Jeremy Howard
- ▶ 27:47 Jeremy Howard So, you know, Jono Whittaker and I, you know, he worked on the stable diffusion course we did.
- ▶ 33:38 Jeremy Howard Like, in stable diffusion.
How to train a Million Context LLM — with Mark Huang of Gradient.ai
- ▶ 38:48 unnamed speaker I think people understand that with stable diffusion, you have these like LoRa patches for like 2 times in the scene
Making Transformers Sing - with Mikey Shulman of Suno
- ▶ 9:42 Alessio Fanelli And then once you get the final model, I would love to learn more about the size of these models because people are confused when stable diffusion is so small.
A Brief History of the Open Source AI Hacker - with Ben Firshman of Replicate
- ▶ 31:53 unnamed speaker And then you moved on to, so was it APIs next, or was it Stable Diffusion next? 2 times in the scene
- ▶ 36:47 Ben Firshman This, like, I think we kind of set up, like, we saw there was this really interesting generative image world going on, so we kind of, you know, like, we were building the tools for that community already, really, and we knew Stable… 4 times in the scene
- ▶ 1:02:58 unnamed speaker And it's like, you need to run like simple diffusion. 2 times in the scene
- ▶ 1:07:29 Ben Firshman Um, and like, like the stable diffusion that we run, for example, is compiled with AI template to make it super fast.
Truly Serverless Infra for AI Engineers - with Erik Bernhardsson of Modal
- ▶ 23:00 Shawn Wang And was that stable diffusion? 4 times in the scene
- ▶ 31:08 Shawn Wang Well, maybe, um, so I'll stick to the stable diffusion thing, and then we'll move on to the other, uh, use cases of, sort of, for AI that you want to highlight.
The Four Wars of the AI Stack - Dec 2023 Recap
- ▶ 41:36 unnamed speaker War, I feel that stable diffusion was like the, the first GPU port model, you know? 3 times in the scene
The Origin and Future of RLHF: the secret ingredient for ChatGPT - with Nathan Lambert
- ▶ 22:49 Nathan Lambert To, like, stable diffusion or whatever.
The AI-First Graphics Editor - with Suhail Doshi of Playground AI
- ▶ 10:25 unnamed speaker It wasn't as big for me as Table to Fusion. 3 times in the scene
- ▶ 17:36 unnamed speaker Um, so you mentioned Dolly, Stable Diffusion. 3 times in the scene
- ▶ 17:54 unnamed speaker I know you supported Stable Diffusion XL in Playground before, right? 2 times in the scene
- ▶ 23:01 unnamed speaker On the one K prompt benchmark, you are 2.5 X prefer to stable the future in Excel.
- ▶ 25:48 Suhail Doshi I think we've talked from, to people, everyone from like Rombok to various people, Rombok trained stable diffusion.
- ▶ 29:14 Suhail Doshi Some old SD, you know, stable diffusion model and be like, look, we beat, you know, stable diffusion, 1.5. 2 times in the scene
- ▶ 29:14 Suhail Doshi Some old SD, you know, stable diffusion model and be like, look, we beat, you know, stable diffusion, 1.5.
- ▶ 29:41 unnamed speaker And, um, you kill stable diffusion, Excel and everything. 2 times in the scene
- ▶ 44:55 Suhail Doshi We kept the playground V two are exactly the same as SD Excel, not because not out of laziness, but just because we wanted, we knew that the community already had tools.
- ▶ 57:12 Suhail Doshi I mean, uh, the guy that, that did Laura, not the guy that invented Laura's, but the person that brought Laura's to stable diffusion, uh, actually works with us, um, on, on some projects.
The State of Silicon and the GPU Poors - with Dylan Patel of SemiAnalysis
- ▶ 29:41 Dylan Patel Open source is doing crazy stuff with stable diffusion, right?
The End of Finetuning — with Jeremy Howard of Fast.ai
- ▶ 4:16 unnamed speaker Fast out of the eye is you teach people from nothing to stable diffusion in, you know, seven weeks or something, and that's amazing.
RWKV: Reinventing RNNs for the Transformer Era
- ▶ 1:07:37 unnamed speaker Um, so, but you could have done the stable diffusion thing. 2 times in the scene
- ▶ 1:42:20 Eugene Cheah And Stable Diffusion literally went through this whole journey.
- ▶ 1:49:46 unnamed speaker I will say, I've actually struggled to find, like, LoRa seems to be very common in the Stable Diffusion community, but in, in text models, I haven't really seen that much adoption in, in my circles, but I, like, maybe you've seen more. 2 times in the scene
- ▶ 1:54:12 Eugene Cheah Uh, because in my opinion, right, like, and this is for me in particular with the stable diffusion model is that if I can get the world image model effectively stable, I mean, stable diffusion was in under a hundred gigabytes. 2 times in the scene
Ep 18: Petaflops to the People — with George Hotz of tinycorp
- ▶ 15:48 George Hotz Um, but I've, I have a rewrite of stable diffusion.
- ▶ 1:01:39 George Hotz And, you know, look, we see the same stuff with all the, like, people are like, stable diffusion's gonna replace artists, or whatever.