Hacker News Reader

Comment by jw1224 | Hacker News Reader

I’ve spent just over 1 week with DALL•E 2.

Over the past 7 days I’ve generated ~1000 images, 150 of which were good enough to save. I only saved images which made me audibly gasp.

Witnessing your own novel idea spring to life is a magical experience. DALL•E provides an artistic tool on a comparable level to digital photography, and by extension Photoshop.

At this stage it’s 100% clear to me that DALL•E has heralded in a revolutionary new age of design. Every day I worked with it, I grew more confident in my outlook.

It might not necessarily be an OpenAI product which truly “integrates” with humanity — but DALL•E has shown me that it’s possible… and just a matter of time.

View on HN

How long did it take for you to generate your images? I've been using https://www.craiyon.com/ for fun but the wait times always results in me getting distracted elsewhere.

I've been using Midjourney [1] (it's not free, 25 photos demo, then 10$ for ~200 photos or 30$ for unlimited I think?). It's fairly fast, ~20s for a grid of 4, then as much for upscaling. I like the controls, it lets you do variations and tweak the image as you go.

It's not as good for doing concrete asks, but it's very good for getting specific vibes.

The website feed [1] requires Discord login to view examples, but there's some unofficial galleries [2]

[0] https://www.midjourney.com/

[1] https://www.midjourney.com/app/feed/all/

[2] https://www.instagram.com/midjourney.gallery/

https://midjourney.gitbook.io/docs/billing#plans monthly plans:

* $10 for ~200 prompts * $30 for ~900 prompts + unlimited "slow" prompts where your job is put at the end of the queue and you have to wait longer (no idea how much longer though.. are we talking about seconds or hours here?)

It probably depends highly on load / time of day. I agree it would be nice to have concrete stats though about the average and variance.

I just got access to Dall-E today, and there it's 115 prompts for 15$, so roughly twice the price.

What's the business with Midjourney demanding Discord details to sign up? Seems off.

The whole thing is run through Discord bot, which is unfortunate but I guess it allowed them to get half their infrastructure for free? You interact with the bot through Discord commands and it generates and returns the photos to you.

It is kind of annoying but lurking and seeing other people's pictures is nearly as enjoyable as making your own. Plus you can use the bot in your dms if you'd like to keep them private.

afaik - the API is exposed as a discord bot, that's how they are managing / restricting access

In a way it makes sense. Discord provides the UI and authentication infrastructure, so the Midjourney devs only have to worry about the actual functionality.

I immediately subscribed to Midjourney after seeing a couple of online tutorials. Prompts/commands being all open source on the community feed was the killer feature for me.

looks like Midjourney's beta invite is expired

The server is open now. Anyone can join. https://discord.gg/midjourney

wow that's one busy server. Seems to be a high demand service.

It's better in some ways than craiyon and worse in others from my sampling (craiyon seems to give me less 'nice' options but better logo-y images)

OpenAI's DALLE UI takes on average <10 seconds to come back with four generations once you submit your prompt.

The quality of results is drastically better than craiyon, but of course you need access and might have to pay for DALL-E. Takes only a few seconds. Also has an editing mode where you can fine tune parts of the image or find variations.

My average generation time for 4 images is around 4 - 8 seconds. It has been much faster than craiyon.

Honestly, I have a great use case for it currently, but then I realized it can only do square pictures, when I really want something that is much wider than it is tall.

You can get DALLE-2 to give different sized images by using their tool to crop out most of the image you already generated, then have it inpaint a completion. You can then use any image editing tool to combine the two images.

I automated the generation of 2048 x 1024 images and started a Twitter bot: https://twitter.com/Bible_or_anime

The relevant code is linked below and is a mess, but the idea is: 1. Generate a base image 2. Use inpainting to expand the left-hand side of the root image. You can do this by submitting an image whose left side is transparent and whose right side is the left side of the root image 3. Ditto for right-hand side 4. Stitch the three separate images together https://github.com/charlesjlee/twitter_dalle2_bot/blob/main/...

In midjourney you can specify the aspect ratio, I think there is an example in their docs

Try MidJourney; 25 free images i think, any aspect ratio

How do you sign up to it? I've found it so confusing, with "join the beta" just being an invite to their discord?

I'd happily pay for it, but have a hard time figuring out how.

Turns out I'm an idiot, there's a sign in with discord button right below it they works just fine.. Join the beta thing threw me off!

Reply on news.ycombinator.com