While we are waiting on DALL*E, you might explore Deep Daze and Big Sleep described here: OpenAI's DALL-E Alternatives with Colab Code - Deep Daze & Big Sleep | GPT-3 for Images - YouTube I’ve been playing with these all afternoon. There are CoLab versions of each.
3 Likes
Waiting on DALL-E, I started playing with #GPT3 + #VQGAN…
@tabarak the monsters and magic items aren’t doing well, but… the spaceships are kinda neat…
Using prompts like…
Drawn in the style of comic book artist Katsuhiro Otomo. The Xim is the name of a large Hive ship that is commanded by a Hive Lord. The ship is armed with many laser turrets that are used to take out enemy ships. The ship is also equipped with a large torpedo that can be launched from the ship.
5 Likes
Really excited for when DALL-E is released! Have lots of ideas and our community at CopyAI has been interested in the developing technology.
1 Like
Thanks. The samples above were made with just a text prompt… I’m playing around to get more control over results…
Draw a fantasy landscape with a big tower on a green grassy plain and a blue sky in the artistic style of Takehiko Inoue
4 Likes
I’d love to generate visual scenes for my favorite literatures or visualizations of the characters, that would be interesting to know how it handles creative inputs
2 Likes
Prompt:
Paint a giant mech battle in the art style of Yohann Schepacz inspired
I love how at the end the AI finally grows the missing leg on the mech on the left… can’t wait for DALL-E!
7 Likes
Literally can’t wait - as a designer DALL-E could help me iterate on concepts much faster… exciting time to be alive!
2 Likes
Yes, we have tons of ideas and needs for high quality synthesized visual assets for our health platform. For example, our app allows users to setup a custom lifestyle/health tracker. It would be nice to generate an image that represents the tracker’s purpose… like tracking my coffee intake… instead of having to provide a large library of assets for the user to choose from.
I know this would impact the stock photo and illustration markets and the artist and photographers that contribute to them… but this is inevitable, the Genie is already out of the lamp.
1 Like
This is one of my faves so far… I used a Bukowski poem as the prompt…
@tabarak, quick question. Will output from DALL-E fall under our copyright like text from GPT-3?
Great question @PaulBellow! I don’t have a concrete answer on that yet. We will be able to share more information closer to a release.
2 Likes
Closer to release… and that would be when? Ballpark 1 month, 3 months , 6 months? Just so excited
2 Likes
Looks really awesome
The general AI community has found a few tricks that improve output quality on CLIP-based image generators (for instance, adding ‘trending on artstation’ or ‘unreal engine’ improves the images dramatically).
I’d be really interested in seeing how well these prompt tricks work with DALL-E.
Prompt: “sam altman OpenAI DALL-E GPT-3”
FacesHQ dataset…
2 Likes
Yeah… I’ve just been experimenting… It knew Bob Ross the painter well. And kinda knows Gary Gygax?
1 Like
Playing with CLIP and SLIP and some other stuff…
I’m about to finish up a 30 second one I did… took around 20 hours this time for 30 seconds…
Any news on DALLE?
1 Like
Hi Paul are you THEE Paul Bellow of LitRPG? - If so just want to say I am a fan
2 Likes
Hah. Thanks.
I realized today I’m using ViTB32 which is the latest from OpenAI, I think?
Latest experiment…
This one has some interesting imagery and almost tells a story? Took 12+ hours to generate. I think I finally know what most of the knobs/dials do, so stay tuned…
This used Katherine Crowson’s fine-tuned 512x512 diffusion model + ViTB32 + RN101
2 Likes