Hacker News Reader

Comment by nl | Hacker News Reader

> If you look closely, these models make horrendous mistakes such as hair growing out of ear, uneven eyes, etc.,

12 months ago I'd have agreed. But these are from raw StableDiffusion, no face enhancement (unless noted). Just created these right now:

https://imgur.com/a/pGPokDq

https://imgur.com/a/yQ10JzB

https://imgur.com/a/dBtVtg3 (Note this has face enhancement on and I censored this so it is SFW)

https://imgur.com/a/TP24KRY

(Still not great at hands yet!)

The rest of your comment is just your misinterpretation of what the models are doing. Modern diffusion models do have large scale knowledge across the whole scene.

View on HN

I mean what I now is that I've got a brand new benchmark for video card upgrades: next time I'm shopping "VRAM to get 1024x1024 resolution of Stable Diffusion" is a dominating factor.

All the current work is a shot across the bow that technology, libraries and research are orbiting some pretty serious utilities.

What does remain to be seen is whether we're in the exponential part of the S-curve or at the top for this particular set of models. But I suspect we'll see the space pretty thoroughly explored in the next 12 months - no large tech company will be taken seriously if it doesn't have a competent image generation model in the wild.

>no large tech company will be taken seriously if it doesn't have a competent image generation model in the wild

Wow, what a lot of bull. Defining a large tech company as 10k+ employees, I can assure you 90% of those won't ever touch image generation. Why would a company like SAP even consider this crap?

If a competitor does it, so will SAP. For instance, if Salesforce can generate meaningful thumbnails for leads or campaigns, etc., then SAP would have to match that.

Assuming this is even a feature users want (they probably don't; why would you have your ERP generate those instead of specialized software?), it makes 0 sense for SAP to build their own model instead of using an open-source one.

Stable Diffusion has been trained at 512x512 and doesn't work very well above this. But upscalers are ok and can even run on CPUs.

They literally lack basic relational understanding so

> knowledge across the whole scene

isn't really correct.

https://arxiv.org/pdf/2208.00005.pdf

Relational understanding is different to whole-scene knowledge.

Relational understanding is about translating the text to the scene.

Whole-scene knowledge is about making sure both eyes are the same color (for example).

Reply on news.ycombinator.com