The Best AI Image Generator Yet

The Best AI Image Generator Yet | Let me Explain

Рет қаралды 1,359

Күн бұрын

Dall-E 3 is about to take the world by storm, with its vastly improved language understanding and prompt coherence. No longer do we need to generate hundreds of seeds. Finally we can get what our dreams desire with this new AI image generator, that also happens to be integrated with ChatGPT!
I would claim this as the first "Real" competitor to Midjourney and a heavily modded A1111. However, keep in mind that Dall-E 3 has plenty of censorship. Hence it will not replace the other tools - but it will be a massive asset.
Here you can find Dall-E 3 updates:
openai.com/dall-e-3
And thanks to Will Depue and Nathan Shipley for the Dall-E 3 examples:
Will Depue: / willdepue
Nathan Shipley: / citizenplain
Here's a link for ChatGPT, where you will be able to use Dall-E 3 if you have ChatGPT Plus or Enterprise (by October 2023):
chat.openai.com/
#dalle3 #AI #chatgpt #midjourney #dalle #A1111
🕒TIMESTAMPS🕒
00:00:00 Intro
00:00:14 The Language Understanding Issue - Solved
00:01:47 Dall-E 3 vs Midjourney
00:03:38 Dall-E 3 Showcase - Domination
00:04:27 ChatGPT with Dall-E 3: The Surfer
00:04:51 Fun ChatGPT + Dall-E 3 Use Cases
00:05:24 Inpainting and Outpainting
00:05:50 Censorship Limitations
00:06:21 What it looks like: Dall-E 3 + ChatGPT
00:06:53 Outro
🔴 Subscribe to check out more of my upcoming content, please! :D
Thank you!
=======================================
🔗 Share this video: • The Best AI Image Gene...
=======================================
SOCIAL
=======================================
Twitter/X: / cakebearcreate
Github: github.com/CakebearCreative

Пікірлер: 20

@BaccarWozat 7 ай бұрын

The censorship is out of control. About half of the prompts are blocked now. Many innocuous things that shouldn't trigger the nono filter, do.

@CakebearCreative 7 ай бұрын

Yeah I feel for you. It’s unfortunate that one of the most powerful models out there must be so censored… These limitations tend to degrade the performance of the image generator too. Let’s hope it gets better!😅

@Vimroth 7 ай бұрын

I like the calm way you explain things. Could you do a brief tutorial on Midas + DaVinci Resolve?

@CakebearCreative 7 ай бұрын

Hello Han! Thanks a lot😁 I’m planning to make a start to end tutorial on making AI movies in the future. I could include some more details about MIDAS, Depth Maps and Davinci Resolve in that one 👍

@Vimroth 7 ай бұрын

Awesome

@markuserikssen 7 ай бұрын

Nice Swedish accent! Dall-E 3 creates impressive stuff.

@CakebearCreative 7 ай бұрын

Hahaha thank you! I guess it’s no surprise I’m Swedish😂 Oh yes, it’s very exciting times for AI!

@FrankysAItime 7 ай бұрын

Nice Video

@CakebearCreative 7 ай бұрын

Thanks a lot! 🐻👍

@linda1541 7 ай бұрын

Dall e 3 in bing images only have 96 dpi. What are the dpi in the paid version? I would like it to be around 300 dpi so I could do some touch ups in Procreate or affinity photo.

@CakebearCreative 6 ай бұрын

Hi Linda! The DPI depends on the physical size of the image. Dall-E 3 images from ChatGPT has 1024x1792 pixels for the 4:7 aspect ratio. For a 24 inch monitor, that would give a DPI of about 92 (~96). However if you e.g. print it on an A4 paper your DPI would be DPI = 1792 / 11.79 = 152. But so most likely, you should have around 96 DPI also with the paid version due to the resolution. I suggest you upscale with an AI upscaler, A1111 is good for this. Then you can reach 300 DPI, if you upscaled ~3x 👍

@Simsonlover222 7 ай бұрын

yes this is what i thought... the understanding of prompts is different between dall e3 and stable diffusion xl... i would love to see a connection between the open source chat gpt (dont know the name, i think lama or something) and stable diffusion.. this would be mindblowing and would elevate stable diffusion even further... stable diffusion evolved so much... but it definetly has its weaknesses... for example i tried fallout 4 power armor... stable diffusion could not give me a power armor that looke like it was from the game.... it was an adaption while dall e3 did exactly that... you saw it and it was fallout 4 power armor :D

@CakebearCreative 7 ай бұрын

Yeah the language understanding of Dall-E 3 is way ahead; it’s a breakthrough! The model is not trained the same way as the other text to img models like Stable Diffusion. I also love Stable Diffusion though, especially with all the open source tools of A1111. But it’s sad that the model is so limited. As you say, maybe one day we get a ”Stable Diffusion Pro” model or something that is trained on GPT3.5 or Llama language. That would be awesome 😁

@Simsonlover222 7 ай бұрын

@@CakebearCreative thx for response :D yeah im pretty sure they will evolve stable diffusion... i remember the beginning of stable diffusion... it was leagues behind dall e... but now you can get the images you cant differenciate if its dall e or stable diffusion... this alone is such a big step up, especially for open source... and then there is google cloud tpu... for a real free open source ai image generator, stable diffusion has no competition... would love to have dall e3 benefits too in stable diffusion... but ey i can generate high quality ai images with amazing results with no effort in seconds and its free... really free... these guys should get a medal... also for not restricting it too much.. and if you run it locally, you can do what the F u want... training models, using different models... the creativity of the community gets together... its like corp vs non corp :D

@Simsonlover222 7 ай бұрын

in my opinion stable diffusion can be used more professionally because of the options it provides.. there are so many options what you can do like changing poses, keep the face but different action and angles... raining own models and using different models... fuse pictures ect. ect.

@Simsonlover222 7 ай бұрын

(if you use it locally, damn i would love to have a fast nvidia GPU... i have a 6950xt but couldnt get it to work effortlessly on my system... would love a simple version for amd)

@brunomazzetti5541 7 ай бұрын

In my opinion, labs is still stuck at dall-e 2

@CakebearCreative 7 ай бұрын

Yeah, currently it is! Will probably take OpenAI a while to fix an inpainting model based on Dall-E 3. But it should come soon 😁

@harrybaker5 7 ай бұрын

incredible tool but the company cant really handle what they have created.... the BING service is really bad probably just due to traffic but the censorship is the real concern, specifically their philosophy behind it. What on earth is wrong with sex and violence? Give the people what they want

@CakebearCreative 7 ай бұрын

I feel your struggle! The censorship is really out of hand. Unfortunately such large companies risk getting sued big-time if they don’t censor😔 However, I actually have an anti-censorship prompt coming up for ChatGPT/Dall-E that helps a lot. Stay tuned 😁👍