Hacker News Reader

Comment by naillo | Hacker News Reader

Is this really using stable diffusion? Has a really 'midjourney' type style to it.

Good eyes! We currently pipe text input to Stable Diffusion, Dall-E and Midjourney at the same time and continue with the most pleasing result. The (very small compared to what we want to achieve) content we currently have required hundreds of renderings and Midjourney came out as most pleasing nearly all of the times. From what's on the website only the Sherlock Holmes renderings have been produced using Stable Diffusion. Nothing published so far from Dall-E, which most of the times is showing us uncanny valley. In terms of preference: We are experimenting hard to achieve better results using Stable Diffusion.

With this effort it sounds it would be faster to actually draw the frames yourself. Get exactly what you want and so on…

One interesting feature of these services is that one can be surprised: suddenly you get an image that is different from what you envisioned when writing the prompt but that is still visually pleasing or interesting or inspiring

Of course, then you have to learn how to draw, which can take years and years for middling results (source: me)

Or in between the two: draw just some blobs to set the general concept and do img2img with diffusion models.

https://simonwillison.net/2022/Aug/29/stable-diffusion/

If the majority of the models were produced with Midjourney, why did you title it as SD as u did?

Probably easier to get on the front page since Stable Diffusion is trending.

The sad thing is that they could’ve also put Midjourney (and DALL-E) in the title and not lost any votes.

Why sad?

>>what we want to achieve

What is it, SPECIFICALLY, you want to achieve.

And, assuming you do achieve THING, what are the negative consequences of such achievement?

How will you be accountable for dark patterns?

What recourse would any biologically physical Human Being have against [THE BAD THING] that may ensue from said achievements...

SERIOUS QUESTION

You literally can't fucking answer this question?

I find stable diffusion requires a lot more prompt engineering than either midjourney or dalle, but has a higher ceiling of quality if you do put in that effort. I wouldn't expect just using exactly the same prompt for all three models would work very well, need at least a little tuning per-model.

Midjourney now used SD underneath

Reply on news.ycombinator.com