DALL-E 2 – Prompt Muse https://promptmuse.com A.I Tutorials, News, Reviews and Community Thu, 26 Oct 2023 09:51:23 +0000 en-US hourly 1 https://promptmuse.com/wp-content/uploads/2022/11/cropped-channels4_profile-32x32.jpeg DALL-E 2 – Prompt Muse https://promptmuse.com 32 32 DALL·E 3.0: Is the Hype Justified? https://promptmuse.com/dall%c2%b7e-3-0-is-the-hype-justified/ Thu, 26 Oct 2023 09:50:15 +0000 https://promptmuse.com/?p=3100 In the ever-evolving landscape of AI-driven creative technology, DALL·E 3.0 is the name on everyone’s lips. With its impressive capabilities, DALL·E 3.0 seemed to outperform its predecessors like Midjourney, and the best part? It was free. But was it too good to be true? Let’s delve into my experience with DALL·E 3.0 to find out. [...]

<p>The post DALL·E 3.0: Is the Hype Justified? first appeared on Prompt Muse.</p>

]]>
In the ever-evolving landscape of AI-driven creative technology, DALL·E 3.0 is the name on everyone’s lips. With its impressive capabilities, DALL·E 3.0 seemed to outperform its predecessors like Midjourney, and the best part? It was free. But was it too good to be true? Let’s delve into my experience with DALL·E 3.0 to find out.

I was one of the early adopters of DALL·E 3.0, and I couldn’t wait to put this new AI image generator to the test. The initial results, I must say, blew me away. What set DALL·E 3.0 apart was its ability to understand not just single words or simple prompts but entire phrases and sentences. It felt like a significant leap forward in AI creativity.

The images it generated were a sight to behold – beautiful, bright, and full of life. It seemed like I could finally do away with the rigid, prompt-based language that had constrained previous AI models. DALL·E 3.0 was a breath of fresh air in the world of creative content generation.

Excitement filled my days as I used DALL·E 3.0 to create a wide array of images for a project I was passionately working on.  It quickly became an indispensable tool in my creative arsenal. However, all good things must come to an end, and in this case, it happened rather abruptly.

One fateful day, I logged onto the platform to create more AI-generated images, only to be met with a disheartening message: “Promptblocked.” I tried again, hoping it was just a minor glitch, but to my dismay, the same message appeared: “Promptblocked.”

A screen shot of a web page showing several pictures related to DALL·E 3.0.
Prompted Blocked  DALL·E 3.0.

It turned out that DALL·E 3.0 had been nerfed into existence. The issue lay in the specific prompt I had used – a celebrity’s name, Lana Del Rey, to be precise. Previously, DALL·E 3.0 had excelled in creating fantastic reference images for my project, but suddenly, it refused to cooperate. The reliable wellspring of creative inspiration had run dry.

A DALL·E 3.0 generated 3d model of a woman with long wavy hair.
A DALL·E 3.0 Lana Del Rey reference Image before DALLE 3.0 got Nerfed.

As I reflected on my journey with DALL·E 3.0, I couldn’t help but acknowledge the rollercoaster of emotions it had taken me on. It had been an incredible tool, enabling boundless creativity with its brilliant images. But, it came with its limitations, and one had to tread carefully to avoid promptblocks.

In the end, my experience with DALL·E 3.0 was indeed remarkable, a brief and brilliant journey through the realm of AI-generated artistry. Though its limitations became evident, the potential of this technology remains undeniably promising. DALL·E 3.0 may have had its ups and downs, but it undeniably showcased the incredible power of AI in the world of creative content generation.

If you wish to try it out for yourself, a link is here: https://www.bing.com/create

It is still a fantastic tool, but not as good as it once was.

PromptMuse

<p>The post DALL·E 3.0: Is the Hype Justified? first appeared on Prompt Muse.</p>

]]>
Consistent AI Characters in any pose https://promptmuse.com/consistent-ai-characters-in-any-pose-written-tutorial/ Thu, 01 Dec 2022 11:39:55 +0000 https://promptmuse.com/?p=541 In this tutorial, we will learn how to train Stable Diffusion with images that currently do not exist. This means you can create any character and train AI to recreate that character in any environment you can imagine. Things you will need: Step 1: Character Design Log into Midjourney. You can use any Text to image [...]

<p>The post Consistent AI Characters in any pose first appeared on Prompt Muse.</p>

]]>
In this tutorial, we will learn how to train Stable Diffusion with images that currently do not exist. This means you can create any character and train AI to recreate that character in any environment you can imagine.

Things you will need:

  1. Google Colab Pro ($8 a month, cancel anytime) https://colab.research.google.com/signup Google colab pro will work as our computer, so you do not need any fancy PC to do this. We will be running this all on virtual machines (It’s super easy!)
  2. Stable Diffusion (AUTOMATIC 111 UI)Automatic 1111 has developed a user interface that can now be installed and run locally on your machine. You need at least 4gb of VRAM to run this, otherwise, you will get out-of-memory errors. But do not fear, I will be bringing a new tutorial that enables you to run stable diffusion remotely without sign up here to find out when it’s released.

Step 1: Character Design

Log into Midjourney. You can use any Text to image generator. It’s just in this tutorial I so happen to be using Midjourney.

https://discord.com/invite/midjourney

You can use any text to image generator you like, I just chose Midjourney as an example. When you log in, find a suitable room on the left-hand side or make your own. In the bar at the bottom type in /settings You can use any of these settings displayed, but again for the purpose of this tutorial, I will be using Midjourney version 4. I switch the Remix feature on as well to get my character close to what I have in mind. See here for more about the remix feature.

My prompt is:

/imagine head and shoulders shot of Instagram model, orange long hair, hyper detailed –v 4

Prompt Muse | A.I News, Tech Reviews and Free Tutorials

My aim is to get the character facing forward. In order to get your character facing forward, you can include in your Prompt looking straight on, looking at camera, symmetrical face.

Do not be afraid to use the remix button to adjust your character.

Step 2: Make a video

The video we need to make is called a Driving video. Think if it like Texas chain saw massacre. We will be taking our characters skin and putting it on top of our animation.

You can use any face (Male/female) to use as driving video, it does not matter. What does matter is that you show an array of facial expressions. Sad, happy, confused, shocked. Also be aware not to turn your head too far left and right, but if you do you can just delete those frames later on- so no biggie.

I used my iphone, you can use webcam or what ever you have to hand.

  • Make sure your video is relatively short (Under 10 secound. Mine was under 20 seconds, and 9.78 mb)
  • Save your video as driving.mp4 & and your characters face image as source.png
  • You can use https://ezgif.com/to crop and resize your video to 401×412 Pixels
  • I matched my characters face and my face in the video up in After effects by just moving it around in place, so the eyes, chin and mouth were roughly in the same place.

Step 3 : Google Colab -Thin Plate Spline Motion

So if this is your first time using google colab, welcome. And if you are a regular user, you probably know more than me. But I think we can all agree to a newbie this looks daunting, it’s not. Use the FAQ and comments section if you have any question about this, I’ll be happy to help.

I advise signing up for the google Pro account, as you will be allocated better GPU and more RAM. Otherwise you may run into out of memory errors. Once you have made an account with google colab.

  • Open the Thin Spline plate motion model notebook and make sure you are signed in.
  • Click on the play button. This will run the cell which installs everything you need remotely on your computer.
  • You will get a message letting you know this is not authored by Google, click Run anyway
Prompt Muse | A.I News, Tech Reviews and Free Tutorials
  • Once a cell has executed successfully you will see a tiny green tick next to the button.
  • We now need to upload our driving.mp4 and source.png.
  • This easiest way to do this is to mount your google drive to this notebook (But I understand for security reasons if you do not want to do this. Please refer to the FAQ here for an alternative method here.
  • Click the files icon to the left of the screen. This will expand you file panel out.
  • Click on Mount drive icon, as seen in image below
Prompt Muse | A.I News, Tech Reviews and Free Tutorials
Prompt Muse | A.I News, Tech Reviews and Free Tutorials
  • Once a cell has executed successfully you will see a tiny green tick next to the button.
  • We now need to upload our driving.mp4 and source.png.
  • This easiest way to do this is to mount your google drive to this notebook (But I understand for security reasons if you do not want to do this. Please refer to the FAQ here for an alternative method here.
  • Click the files icon to the left of the screen. This will expand you file panel out.
  • Click on Mount drive icon, as seen in image below
Prompt Muse | A.I News, Tech Reviews and Free Tutorials
  • Once you have clicked the Mount drive icon, a new cell will appear in your code section of you notebook. Click run (The play icon)
  • You will now be asked to connect your google drive to this notebook. Click connect to google drive, and log into your google drive.
  • Once this cell has sucessfulyl excuted, you should now see a new file appear in your files panel on the left hand side (Might take a few secound to appear. If not you can press the go up a folder icon, this will refresh your folder list. Now navigate to: Content > Thin-plate-spline-motion>Drive
Prompt Muse | A.I News, Tech Reviews and Free Tutorials
  • Now, go back to you google drive in Drag and drop your driving.mp4 and source.png into your google drive folder making sure it’s not in a folder. Right click on each file and click Get link. On the general access drop down select Anyone with Link and then Done.
Prompt Muse | A.I News, Tech Reviews and Free Tutorials
  • Navigate back to your Thin Plate Spline notebook and Right click on your driving.mp4 (Located in the left hand file structure) and click Copy Path

Paste the path into Step2 settings,

source_image_path:
/content/drive/MyDrive/source.png

driving_video_path:
/content/drive/MyDrive

Run cell Step 2

Prompt Muse | A.I News, Tech Reviews and Free Tutorials
Prompt Muse | A.I News, Tech Reviews and Free Tutorials
  • The next steps are easy, just run the cells in order and wait for them to complete before moving onto the next cell
  • Once all cells are complete you will have all you assets that you have created saved in the folder structure on the left hand side. All you now need to do is download the up scaled frames. Save about approx 20 of the best frames, showing various facial movements.
Prompt Muse | A.I News, Tech Reviews and Free Tutorials

Step 4 : Outpainting

Hey, I’m just finishing this documentation up, so bare with me whilst I work on it. The rest of the tutorial Should be up within 24hours 

<p>The post Consistent AI Characters in any pose first appeared on Prompt Muse.</p>

]]>
How to fix AI art faces : Midjourney, Stable Diffusion, Dalle 2 https://promptmuse.com/how-to-fix-ai-art-faces-midjourney-stable-diffusion-dalle-2/ Wed, 30 Nov 2022 20:03:41 +0000 https://promptmuse.com/?p=238 Have you ever taken a mid-journey photo using AI-generated art only to find your image looking like a wonky mess? Fear not, we’ve got you covered. In this guide, we’ll show you three methods to fix those facial features, smooth out skin and achieve a more harmonious image. Method One: Arc by Tencent Arc by [...]

<p>The post How to fix AI art faces : Midjourney, Stable Diffusion, Dalle 2 first appeared on Prompt Muse.</p>

]]>

Have you ever taken a mid-journey photo using AI-generated art only to find your image looking like a wonky mess? Fear not, we’ve got you covered. In this guide, we’ll show you three methods to fix those facial features, smooth out skin and achieve a more harmonious image.

Method One: Arc by Tencent

Arc by Tencent is a simple and free inner browser app that can be used without an account or any registration. Follow these steps to use it:

  1. Navigate to Arc and hit the “Upload” button.
  2. Select the AI-generated image that you want to fix from your files. The upload process may take some time to complete.
  3. Use the app’s “before and after” feature to adjust the settings until you’re happy with the outcome.
  4. Click the “Download” button to retrieve the finished image.

Note: This method may not be suitable for illustrative or textured images, as it makes them appear photorealistic.

Method Two: Gfpgon

Gfpgon is a program that can be run on Google Collab. It’s free to use and can be accessed via your Google Drive account. Here’s how to use it:

  1. Go to Gfpgon and click on “Connect”.
  2. Complete each of the five steps by clicking the “Play” button next to each one.
  3. Wait for the upload to complete.
  4. Click on “Visualise” to see the finished image.
  5. Click on “Download Results” to download the final image.

Note: This method is slightly more complicated and requires a Google account to use. However, it produces high-quality results.

Method Three: Using Photoshop

If you’re a Photoshop user, this method may be the most familiar to you. However, it’s also the most time-consuming and requires a subscription. Here’s how to use Photoshop to fix your AI-generated images:

  1. Open the image in Photoshop.
  2. Use the brush tool to paint over the facial features that need fixing.
  3. Adjust the brush size and opacity as necessary to get the desired effect.
  4. Save the image and compare it to the original.

Note: This method provides the most control over the final image, but may not be the most accessible for everyone.

So there you have it, three methods to help fix those mid-journey AI-generated images. Whether you use Arc by Tencent, Gfpgon, or Photoshop, you’ll be able to achieve a more harmonious image in no time. Experiment with these methods to see which one works best for you. Let us know in the comments which one you prefer!

 

Transcript

 

Hello and welcome back to another prompt news video. Today I’m going to be showing you how to fix those midjourney junky faces. This also works for darley output images and stable defusion. Well, essentially any AI generated art. So come along and I’ll show you how to fix those with wonky eyes and funny noses.

So, the first method we are going to be using is Arc by Tencent. It’s an inner browser app and you don’t need to log in, you don’t need to make an account, and you certainly don’t need to pay for anything. So it’s really easy. And let me show you how to use it. So we navigate over to Arc and as you can see, it’s pretty much of a WYSIWYG.

So we’re going to hit upload and find my Janky folder. This image was actually created in midjourney, if you wanted to know. I can’t remember the prompt. It was something about girls flashing and scene. So it takes about 30 seconds for it to load your image in.

Unfortunately, sometimes these in browser apps can be a bit slow because there’s a lot of people using the service. Here we go. And what I like about art, it shows you the before and the after. So you get this little scrolling thing and you can just push it back and forward. And to see the difference that the app is making here now, you can see it’s not only sorting out the facial features, it’s smoothing the skin and giving a colour correction as well.

And I’ve flipped right over. It’s actually getting a ridiculous bit of the shininess on the nose and refining some of the environmental detail. Now, I think that looks quite good. The only thing is that it’s moving and removing any of the textures. So if you’ve got kind of a textured illustrative look, it might not be the programme for you because it’s going to make it kind of photorealistic.

But if you want a quick fix to upload your images to the Instagrams, this is a very quick and easy process. And you just click on download and the job is done and it spits out your image for you. Okay, now to method two. This is called gfpgon, and it’s run on Google collapse. Please don’t be intimidated by the user interface.

It’s very easy to use. What’s really cool about Gspg is that you can actually save and run it from your Google Drive and make modifications to the app. All you need is a Google account. But for now, I’m not going to be doing that. I’m just going to click Connect here and then that will connect you.

There are five steps in total and you just click the ticks next to them and upload your image. So no talking. Let’s get to it. So we scroll down to step one, preparation. The user interface is showing user executions being run.

But don’t worry about that, you don’t need it. You’re not a programmer. So when you hit the Play button here, what it will do is run the code and you just wait to the bottom until it says 100%. It usually takes about 30 seconds. In fact, it tells you how many seconds it takes.

We’re 100%. Okay, so let’s move on to upload images. And that’s number two. So again, hit the Play button. Now you can select your image from your computer.

So I’m going to go to browse, get my Janky image. It’s important to wait until the Play button has a green text next to it and then you can move on to the next step. So it just takes a minute. I’m just going to have a cup of tea. One thing is to note you can see the progress of your image being uploaded here at the bottom, so you’re not waiting for an infinite amount of time.

Okay, that has now loaded. I’ve got my green tick. Let’s move on to inferring. So find that Play button and hit the Play button again at the bottom. Here we have the progress percentage, so keep an eye on that.

It usually takes up to 40 seconds, so it won’t take long at all. OK, so the next step is to visualise your image. So click the play button. Once again, we’ve only got one step after this. If you scroll down now, you will see your images.

And again, it’s got the comparison between what you put in and what it fits out. So it has a very similar post effect as Arc does. As you can see, it’s created a symmetrical image, a more harmonious image. It has sharpened the image and given a smooth to detected skin, as well as upscale the image slightly. And then the fifth and final step is to hit Play on number five download results and that will download you a lovely zip out of your image.

So our third and final text legal method is using Photoshop. You will require a subscription for this, so it’s not free and you need some skills. So with a quick tutorial on YouTube, you’ll be able to paint the tie no problem, I’m sure. But this is the final technique and I’ve done a really rubbish, don’t judge me by this, but a very quick eye repaint so you can see what you can achieve. Now, personally, I prefer this method out of the frame.

You can create your eye from scratch and ultimately have complete artistic control over your design. Also, you keep the integrity of the original painting. So if it’s done in quite an illustrative style or a loose painting style, you can keep those details. And here is a comparison of the faces across the applications. I’ve got the original, then Arc and then GFP gon, and then two minutes in Photoshop without any colour correction.

So Arc and Gfpg are actually pretty much photorestoration applications, but you can use them to fix up your dodgy AI. I would probably suggest investing some time and knowledge in getting to learn photoshop. Then you can become a master of your own destiny. So that is the free method, the predict of unjunctrifying AI images. If you have a better method, let me know in the comments below and I’ll try those out.

So thank you very much and that will do it. Bye bye.

Oh, yeah. Oh yeah. Always forget this bit. Can you like and subscribe and hit the notification button and I will come out with all video. Okay, thanks.

Bye.

<p>The post How to fix AI art faces : Midjourney, Stable Diffusion, Dalle 2 first appeared on Prompt Muse.</p>

]]>
How To Write Prompts  https://promptmuse.com/how-to-write-prompts/ Wed, 30 Nov 2022 19:57:49 +0000 https://promptmuse.com/?p=227 Welcome back to another prompt news video. I am so happy to have you guys here and thank you for the new subscribers. As always, if you want to subscribe and hit the notification bell, that makes all this worthwhile. In today’s video, we are going to be discussing the best ways to write prompts [...]

<p>The post How To Write Prompts  first appeared on Prompt Muse.</p>

]]>
Welcome back to another prompt news video. I am so happy to have you guys here and thank you for the new subscribers. As always, if you want to subscribe and hit the notification bell, that makes all this worthwhile. In today’s video, we are going to be discussing the best ways to write prompts and some of the bad ways to write. This video is for beginners. So but also if you’ve been using it for a while and you’re a bit of a dab ham, this might give you some tips and advanced tricks that you can integrate into your pumps to get a better image and better result. So as I filmed this video, AI is still pretty new and you can find it on multiple platforms. And today I’m going to be using Mythjourney. It’s important to understand Mythjourney is still in developmental space. All this AI is pretty new to the scene. So you will see glitches, you might see a person with seven arms, or you might be having issues with hands. And that’s a classic of journey issues that they’re working on. But the faces are absolutely stunning. So using my journey as a concepting tool is fantastic.

You can try out different subjects, you can try out different styles, cyberpunk, steampunk, art nouveau, anything you can think of and add different artists to the mix and different mediums and styles and lighting. It blows my mind every time I use it and it’s only going to get better. This is an exciting time to come on board and learn how to write prompts. If you’re thinking, oh, this sounds very complicated, I don’t know where to start, start here. Literally, the secret is writing prompts is the same as writing. In Google Search, you are writing a text description of what you want to see and then hitting return and then midjourney. It brings you back an image result that it thinks you want to see based on your text description, exactly like a Google Search. So I’m going to break the prompt down into four simple, understandable components. So first you’ve got your concept and your subject. This is where the most weight is going to be at the beginning of the font. This is where you will define what it is. The next section is style. So you can style your artwork with a genre or a particular artist or mix in different artists.

You can use multiple artists and or different medium of art. So you could try photography, charcoal, sculpture. Just have a play around with all these different styles and you’ll be able to come out with some pretty cool images. So next is the quality inputs. HD stands for high Definition. You can add cinematic lighting. You may have seen Octane Render, which is a 3D renderer. The last and final part of your prompt is the composition. So how you want your output to look like. Do you want an Ultrawide shot? Do you want a wallpaper, like a desktop wallpaper. Do you want a macro shot or a specific headshot of your concept or subject? Put that in here. You can also put your aspect ratio so you can add the ratio size or the pixel size that you want your image. This will change your composition and sometimes gives you better images. So play around with the aspect ratio. Now, if there’s any words that you don’t understand on this screen, google it. Get used to all these different terminologies AI.

Art is not an exact science.

You can become a director and by using the correct prompts, you can get closer to the feeling or the vibe that you want from the image. So all of that in practical terms, it’s all well and good. So let’s put the theory into Practise. Let’s type in sitting black hat, wearing glasses, art by Art Germ, cute cinematic style, HD detail, octane render, and the aspect ratio of two by three. This is the result of our prompt. I did a couple of variations on this image and when I was happy with the variation, I upscaled the image and this was the result. And I think you can agree it’s pulling all the information from the prompt. It’s a black sitting cat with glasses on. It’s cute, it’s got a cinematic style, it referenced the artist nicely, and the aspect ratio is two by three. In this circumstance, Mid Journey has understood the brief and I know it’s not always a way. And you might have to reroll your gift closer to what you want. So you might be wondering what would happen if I just took all the fancy pants descriptions out and just put black cat in glasses.

Well, this is what my journey would give me. It is a black cat in glasses, but there’s no stylization and it quite literally is a black cat in glasses. So you can see how the prompt really influences the output that midjourney will retrieve for you. So hopefully, from this point forward, by watching this video, your prompt structure has improved. Now, to improve on your language, you can go and get inspired by heading over to the Midjourney.com website. And if you go below home and see Community Feed over on the left and click on that, you can see what images are trending, what images are popular, what images are new. What’s really cool is you’re on the standard subscription. If you click on any image and scroll down, ignore this text here, that’s not the prompt. If you click on the two dotted lines, three dotted lines here, and click on Copy Command and open Notepad, Word, whatever you got and just paste, you can see the exact command that they used for that image. Now, if you put this into Midgenit, you will not get a same image because it’s re rolling the dice. It’s not the same seed number, which I can explain in another video, but you will not get the same image, but you’ll get something similar.

But you can see that they have used pretty much all the structures that I explained earlier on in the video. So they put their main concept subject. Their artists actually use the same artist reference and what kind of design, what kind of quality they want from the image. As well, you may notice no watermarks. So if you want to subtract something from your prompt, you put in no glasses and that would remove the glasses from my cat. Or if you were doing a landscape, I’ve been hyping no trees and that would remove all the trees from your image. So that is a pretty cool trick. They also don’t want any borders, they don’t actually want any signature on their image. As you notice, some images from midjourney will show them swinging line a signature and you usually have to photoshop those out. But if you request no signatures, your image won’t have that in. That’s a really cool thing to learn as well. So, as I said, when you come over to the midjourney community, you’ve learned so much by looking at other people’s artworks. So we spoke about all the good things to put into your prompt and how to structure them.

There are things that you should not include in your prompt and those are banned words. You can go to the Journey Discord server and on there you can find the Rules channel. And on the channel the statement goes as follows do not create images or.

Use text fonts that are inherently disrespectful.

Aggressive or otherwise abusive. Violence or harassment of any kind will not be tolerated. No adult content or gore. Please avoid making visually shocking or disturbing content. We will block some text inputs automatically. So there you go. That’s everything you need to know in a nutshell. But if you are still unsure, just head over to the Rules section on the Discord Server and you will find their terms and conditions. I try and keep these videos as bitesized as possible because I know it is a lot to take in and I do have other videos that expand further on prompts, but I hope of every video I do, I improve and get you the information that you need clearly and concisely. I would really, really appreciate it if you follow my channel and subscribe.

And just before we go, I don’t want to say I have a prompt music Instagram page where we can interact over there and you can see prompts and images that I post. I pick a muse a day, so it’s worth following. And if there are any tips you feel I’ve missed out, please feel free to put in the comments section so other people can see and learn. Thank you so much guys, for joining me for another video and I’m looking forward to making another one. And that’s it for now.

Thanks a lot. Bye.

<p>The post How To Write Prompts  first appeared on Prompt Muse.</p>

]]>