Jul 12, 2022·1 reply
Stay HealthyWhat is the explanation for the difficulty with eyes and faces. Words being odd makes sense to me though
Also is any source code being released.
with Dall e, idk. possibly the disproportionate amount of unique detail in a relatively small area compared to other objects/concepts, which would explain why starting with close ups works v well
There’s also the heavy filtering particularly of faces from the dataset used for training the model
Check OpenAI’s blog posts about the training of the model. There’s also the model card on GitHub and links to papers if you’re interested enough
Jul 12, 2022·1 replynOVOhahagottem· editedxi tried for a while but just couldnt get a good one tbh. for some reason it just couldnt make the monocle right. also out of images for today.
Bro these are amazing 🥺
Can you teach me how to make them better?
Here’s one I made
- Jul 12, 2022·1 replynOVOhahagottem· edited
xi tried for a while but just couldnt get a good one tbh. for some reason it just couldnt make the monocle right. also out of images for today.
Scott is it possible to get a super upscaled version of the 1st pic?
I’m gonna print it out and hang it on my wall lol
OPJul 12, 2022
no fair
need them to send my invitation- Jul 12, 2022·1 replySquilliam
Scott is it possible to get a super upscaled version of the 1st pic?
I’m gonna print it out and hang it on my wall lol
it should be 1024x1024 if you download the full image
best way to upscale would be to use an upscaling ai like esrgan i think?
- Jul 13, 2022nOVOhahagottem
it should be 1024x1024 if you download the full image
best way to upscale would be to use an upscaling ai like esrgan i think?
Ohh great thank you. I know in the discord there’s a button for upscaling but I believe you’re using a different program
Jul 13, 2022that's cray and all but where is the p*** ai?
Jul 13, 2022·1 replyGot my DE2 invite
lil ufo 🛸Jul 13, 2022·2 replieswe made it to highsnobiety instagram.com/p/Cf7EwY8M3Hk/?igshid=YmMyMTA2M2Y=
- Jul 13, 2022·1 reply
dall-e2 invites going out like crazy
midjourney launched public beta
- OPJul 13, 2022lil ufo
we made it to highsnobiety https://www.instagram.com/p/Cf7EwY8M3Hk/?igshid=YmMyMTA2M2Y=
- OPJul 13, 2022
just signed up for the dall e 2 waitlist on my other email just to see if i can get it on that one 
- OPJul 13, 2022wheatley
Got my DE2 invite
show us some art 
Jul 13, 2022·1 replygaretarecredit @BGFX
i think this is the best one by far tbh
- garetare ✅Jul 13, 2022·1 replyBGFX
i think this is the best one by far tbh
Deidirestere
- Jul 13, 2022garetare
Deidirestere
Blanquito Mierda 🖕🏻Jul 13, 2022·1 replynOVOhahagottemdall-e2 invites going out like crazy
midjourney launched public beta
Nothing in my email.
- lil ufo 🛸Jul 13, 2022Blanquito Mierda
Nothing in my email.
just joined the list with all my e-mails
- Jul 13, 2022·edited·2 repliesSquilliam
Bro these are amazing 🥺
Can you teach me how to make them better?
Here’s one I made
honestly those really werent that difficult w/ dalle2. just did several prompt variations
ill link some resources on prompt design
with midjourney its all about iterating and iterating with variations and prompts a LOT. choose the best variations each time, look closely. etc.
also read the docs. i always use a width of something like 360, 420 or 500
e.g.
A highly detailed digital 3D render of a black cat wearing a gold chain, trending on artstation, character design by Pixar --w 420`tried it just now and had to go through like 40-50 variations to get something that didnt look all messed up
mj isn't as good as dall-e at understanding natural language and how things connect with each other. i mean its still good, but not compared to dall-e's language understanding abilities
With DALL-E you get more coherent results immediately. but the prompts that work well with MJ dont work well with dall-e and vice versa. you can do better than the above images with less effort on mj but im rusty w/ mj prompt design after using dall-e so much
Jul 13, 2022·1 replynOVOhahagottemwith Dall e, idk. possibly the disproportionate amount of unique detail in a relatively small area compared to other objects/concepts, which would explain why starting with close ups works v well
There’s also the heavy filtering particularly of faces from the dataset used for training the model
Check OpenAI’s blog posts about the training of the model. There’s also the model card on GitHub and links to papers if you’re interested enough
@sco is this part of your profession/career field or just a hobby?
- Jul 13, 2022·1 replyPhlegm
@​sco is this part of your profession/career field or just a hobby?
yo, the art stuff i do for fun
AI in general i've been learning about over the past 18 months as part of developing this platform. plus general interest + potential future opportunities
you in the creative industry?
- Jul 13, 2022·1 replynOVOhahagottem
yo, the art stuff i do for fun
AI in general i've been learning about over the past 18 months as part of developing this platform. plus general interest + potential future opportunities
you in the creative industry?
half in half out
been using ai in my art and playing/researching with it for projects for a few years now. mostly through reading the older lit and playing with collab notebooks - Jul 13, 2022·1 replyPhlegm
half in half out
been using ai in my art and playing/researching with it for projects for a few years now. mostly through reading the older lit and playing with collab notebooksyea i started out with notebooks last year. do you have mj or dall-e access?
- Jul 13, 2022·1 replynOVOhahagottem· edited
honestly those really werent that difficult w/ dalle2. just did several prompt variations
ill link some resources on prompt design
with midjourney its all about iterating and iterating with variations and prompts a LOT. choose the best variations each time, look closely. etc.
also read the docs. i always use a width of something like 360, 420 or 500
e.g.
A highly detailed digital 3D render of a black cat wearing a gold chain, trending on artstation, character design by Pixar --w 420`tried it just now and had to go through like 40-50 variations to get something that didnt look all messed up
mj isn't as good as dall-e at understanding natural language and how things connect with each other. i mean its still good, but not compared to dall-e's language understanding abilities
With DALL-E you get more coherent results immediately. but the prompts that work well with MJ dont work well with dall-e and vice versa. you can do better than the above images with less effort on mj but im rusty w/ mj prompt design after using dall-e so much
