Dall-E, Midjourney, Stable Diffusion (etc) - who's playing, and with which?
Dall-E, Midjourney, Stable Diffusion (etc) - who's playing, and with which?
- Started
- Last post
- 842 Responses
- PhanLo0
AR skeleton filmed inside a real environment converted to a guassian splat. (mindblown . gif)
https://lumalabs.ai/capture/3E56…
-- insane!
how?!microkorg - Was just using the Luma app on the phone Micro. You record a video then it generates the scene. https://lumalabs.ai/…PhanLo
- damn, dont have iphonemicrokorg
- I think it can work through the website too maybe?PhanLo
- insane!
- microkorg1
Got access to https://magnific.ai/
It is Magnifique!!- I like how in the leaf example it completely changes the species, lolNairn
- It is quite unpredictable. When it doesnt understand something - like abstract art or theres not something recognisablemicrokorg
- then it'll struggle like other AI upscalers etc. But when it comes to photographic or illustrative content its astounding.microkorg
- doesn't always look better in the after. just more detailed. Can you not just do this in mid journey? upload an image and prompt it to enhance?_niko
- this is upscaling and enhancing.
still playing with it. im 100% sure its useful for me and my art style. MJ & Topaz are good enough for me at the mo really.microkorg - Topaz GigaPixel AI works greatutopian
- Looks like they're just using diffusion with lower denoise, also probably some kind of clip retrieval for prompting.kalkal
- (conditioning)kalkal
- _niko1
- It's sorta like a graphic designer using a typeface, adding a stock photo and then having a programmer or printer execute the idea that's not yours.utopian
- myeah, i don't buy this approach - you're 'making' something digestible to people who don't care enough to over-analyse things. Just, do it well.Nairn
- also, prompts are *hard*. Maybe i'm just a bit thick, but it takes a fair amount of effort to get what I want.Nairn
- Nairn2
Visual Anagrams: Generating Multi-View Optical Illusions with Diffusion Models
tl;dr: We use pretrained diffusion models
to make optical illusions- those are quite badhans_glib
- https://i.imgur.com/…
there are a few mind-blowing ones. these are the same image invertedkingsteven - or the jigsaw that draws Einstein or a potted plant depending how you arrange the pieces...kingsteven
- PhanLo0
- nice, they released the source. colab here https://github.com/c…kingsteven
- ha, they have their own hugging face demo now too. love how quick this stuff moves (even after wasting hours setting up one myself)kingsteven
- it does seem totally useless though haha. did not get one decent resultkingsteven
- Have you got a link to that? I couldn't find itPhanLo
- https://huggingface.…kingsteven
- i'm guessing a bit of competition going on with the AnimateEveryone project you posted. It seems like a better implementation but this is good for a laughkingsteven
- ^ Cheers man, I've set something to render so will se what happens. Took a while to get a spot.PhanLo
- PhanLo3
- A bit o'a fan o'a bit o'pipe are ye, Phanlo?
:)
This is great - pure prompt, or derived from some sort of source?Nairn - mad for pipes, laying pipe all day long.
I used this as the base https://i.imgur.com/… then used controlnet to fancy it upPhanLo - It was the same prompt once through a fast SDXL model on Huggingface then through the Alkaid plugin in Photoshop.PhanLo
- Hey Mario....in your past life you were a plumber.utopian
- I plumbed in a couple of taps in the house a while back, wasn't as fun as AI generating pipes or drawing them, that's for sure :-)PhanLo
- A bit o'a fan o'a bit o'pipe are ye, Phanlo?
- PhanLo2
- Dig itYakuZoku
- I know you dislike the concept of NFTs PhanLo but you could be upscaling these and selling all this work. Far better than a lot of the AI work that sells.microkorg
- I just can't wrap my head around NFT's.
Love your stuff, really love it, but it seemed like such a hassle to set up selling anything.
Seemed worse thn galleriesPhanLo - Do you still make money from them?PhanLo
- This year has been pretty crap for sales. But it has been good for marketing/exhibiting though.
Tokyo x2, NY x2, London x2.microkorg - Branching out and doing more physical stuff in 2024. Prints, decks and likely a big pricey book :)microkorg
- Will still do NFTs. Got a WIP generative project. And want to do some blind-buy series where not revealed to end and do it like a raffle.microkorg
- Where some of the pieces will be a print or deck to be won.microkorg
- ETH is a bit crazy with minting prices. Think you'd vibe with Tezos on objkt.com
It costs pennies to mint.microkorg