Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Stable Diffusion 2.0

stability.ai

Post reply on HN
Print view View original

281–290 of 519 posts

Re: Stable Diffusion 2.0

#281

Post by nicoco »

acidburnNSA wrote: post #156

Awesome. I'm installing on Ubuntu 22.04 right now. Ran into a few errors with the default instructions related to CUDA version mismatches with my nvidia driver. Now I'm trying without conda at all. Made a venv. I upgraded to the latest that Ubuntu provides and then downloaded and installed the appropriate CUDA from [1]. That got me farther. Then ran into the fact that the xformers binaries I had in my earlier attempt…

Also got this far on my 3080 Ti with the same error message. Oh well, let's wait for the "optimized" forks to pop up.

Re: Stable Diffusion 2.0

#282

Post by myrloc »

prakhar897 wrote: post #172
LASR wrote: post #27

I am a solo dev working on a creative content creation app to leverage the latest developments in AI. Demoing even the v1 of stable diffusion to the non-technical general users blows them away completely. Now that v2 is here, it’s clear we’re not able to keep pace in developing products to take advantage of it. The general public still is blown away by autosuggest in mobile OS keyboards. Very few really know how far…

I'm in a kind of same boat. I think indie games are the way to show true potential of SD. Hence, I'm working on http://diffudle.com/ which is a mix of Wheel Of Fortune + Stable Diffusion + Wordle. I Can't figure it out but feels to me like its lacking something.

I really want to try this. Please add mobile iOS support!

Re: Stable Diffusion 2.0

#284

Post by ShamelessC »

millimeterman wrote: post #161
liuliu wrote: post #41

Seems the structure of UNet hasn't changed other than the text encoder input (768 to 1024). The biggest change is on the text encoder, switched from ViT-L14 to ViT-H14 and fine-tuned based on https://arxiv.org/pdf/2109.01903.pdf . Seems the 768-v model, if used properly, can substantially speed-up the generation, but not exactly sure yet. Seems straightforward to switch to 512-base model for my app next week.

I'm disappointed they didn't push parameter count higher, but I suppose they want to maintain the ability to run on older/lower end consumer GPUs. Unfortunately it severely limits how high-quality the output can be.

Can’t forget time it takes to run inference, even on the latest A100/H100. Generating in under e.g. ten seconds enables more use cases (and so on until high fps video is possible).

Re: Stable Diffusion 2.0

#285

Post by thot_experiment »

TaylorAlexander wrote: post #154
kmeisthax wrote: post #58

Is there a good explanation of how to train this from scratch with a custom dataset[0]? I've been looking around the documentation on Huggingface, but all I could find was either how to train unconditional U-Nets[1], or how to use the pretrained Stable Diffusion model to process image prompts (which I already know how to do). Writing a training loop for CLIP manually wound up with me banging against all sorts of stra…

Ah I am glad to see someone else talking about using public domain images! Honestly it baffles me that in all this discussion, I rarely see people discussing how to do this with appropriately licensed images. There are some pretty large datasets out there of public images, and doing so might even help encourage more people to contribute to open datasets. Also if the big ML companies HAD to use open images, they would…

The problem is not that people aren't owning ideas hard enough, ideas shouldn't be ownable in this way, the problem is that we've created a system that's obsessed with scarcity and collecting rents. Being able to own and trade ideas a la copyright/patents helps people who can buy copyrights and patents stifle creativity more than it helps artists gather reward for their creation (though it does both).

Human endeavor is inherently collaborative. The idea that my art is my virgin creation is an illusion perpetuated by capitalists. My art is the work of thousands who came before me with my slight additions and tweaks.

Your (and in general, our) suggestion that we should be concerned with respecting or even expanding these protections is incorrect if you want human creativity to flourish.

Re: Stable Diffusion 2.0

#286

Post by UweSchmidt »

prakhar897 wrote: post #172
LASR wrote: post #27

I am a solo dev working on a creative content creation app to leverage the latest developments in AI. Demoing even the v1 of stable diffusion to the non-technical general users blows them away completely. Now that v2 is here, it’s clear we’re not able to keep pace in developing products to take advantage of it. The general public still is blown away by autosuggest in mobile OS keyboards. Very few really know how far…

I'm in a kind of same boat. I think indie games are the way to show true potential of SD. Hence, I'm working on http://diffudle.com/ which is a mix of Wheel Of Fortune + Stable Diffusion + Wordle. I Can't figure it out but feels to me like its lacking something.

Very creative and a fun way to interact with SD. I would encourage you to explore this idea further, as interest in SD might grow and people want to engage with the topic in an accessible way. I like the idea of hard-limiting play (1 quizz per day) but a small backlog of previous pictures could be nice to explore a little.

Re: Stable Diffusion 2.0

#287

Post by Naracion »

myrloc wrote: post #280

“Adoption” is a generous term to use for a description of Github stars (referring to the first graph). There’s no denying stable diffusion has been gaining popularity, but I think it’s hard to say it’s really being adopted at the same rate it’s getting starred on Github.

FWIW, they only call it "Github stars". The graph that says "adoption" is from a16z [0].

0: https://a16z.com/2022/11/16/creativity-as-an-app/

Re: Stable Diffusion 2.0

#288

Post by Tepix »

acidburnNSA wrote: post #156

Awesome. I'm installing on Ubuntu 22.04 right now. Ran into a few errors with the default instructions related to CUDA version mismatches with my nvidia driver. Now I'm trying without conda at all. Made a venv. I upgraded to the latest that Ubuntu provides and then downloaded and installed the appropriate CUDA from [1]. That got me farther. Then ran into the fact that the xformers binaries I had in my earlier attempt…

Which GPU are you using? Used RTX 3090s were relatively cheap in the last couple of weeks...

Re: Stable Diffusion 2.0

#289

Post by ShamelessC »

eoerl wrote: post #179
corysama wrote: post #170

Earlier quoted context omitted.

It’s my understanding that, amazingly enough, blending the models is done by literally performing a trivial linear blend of the raw numbers in the model files. Someone even figured out they could get great compression of specialized model files by first subtracting the base model from the specialized model (using plain arithmetic) before zipping it. Of course, you need the same base file handy when you go to reverse…

It is not typically possible to blend models like that, since the training process is (lateral) order insensitive, as far as the model goes.

Is that still the case when all models have a common ancestor (i.e. finetuned) and haven’t yet overfit on new data?

Re: Stable Diffusion 2.0

#290

Post by rrobukef »

Jack000 wrote: post #141
hackater wrote: post #130

Earlier quoted context omitted.

Can you go into a bit more detail? What architecture did you use? Is the month training time really just training with mini batches with a constant learning rate? Or are these many failed attempts until you trained a successful model for a few days in the end? I particularly interested in the image generation part (the DDPM/SGM)

Yeah I did have a few false starts. Total time is more like 3 months vs 1 month for the final model. For small scale training I found it’s necessary to use a long lr warmup period, followed by constant lr. There’s code on my GitHub (glid3) edit: The architecture is identical to SD except I trained on 256px images with cosine noise schedule instead of linear. Using the cosine schedule makes the unet converge faster bu…

I keep wondering if using not only statistical noise but also deformations would help with the generation of deformable things - say human hands.
Post reply on HN