Stuff is so crazy. 😭 Been able to build a personal brand with just an android device and $0 startup. Won some giveaways and made some awesome connections. Can't imagine what is possible with a laptop, chatgpt and these new tools like avatar Creation, image animation and more. Completely mind blown with Dall e 3 quality.
I have been empowered by all of your amazing videos. I wonder if you know a way to create consistent looking characters in DALL-E 3 within ChatGPT. I wonder if there’s a way to use a particular image of a character and adjust it so that it looks the same in each image, but is changed sequentially from page to page. For example, a character for a children’s book that meanders through a story, and looks identical in every image, even though they are performing a different task throughout the story. In other words, it’s the same person throughout the book and you recognize that it is the same person or main character such as a cat or a dog. Cheers and thank you. Sincerely.
Thanks! I have explored this a little. Because chatgpt can remember, consistent characters should be easier. My results have been hit-and-miss. Sometimes I think ChatGPT has it figured out. Other times it doesn't. Let me play with this a little more. You should - in theory - be able to describe the character and ask ChatGPT to keep features the same. The character probably won't be EXACT, but close.
I don't think the tool is that precise right now. Ask for a tall image. "Make a tall image of a -----" You'll get something like 9:16. I did successfully get the AI to put a frame around the image for other aspect ratios, but it's hit-or-miss.
I've had a plus account as long as they've existed. ChatGPT is basically my PA these days. If I could plug it into my coffee maker, it would also bring me coffee in the morning.
Amazing tips and way of presentation. I love that you highlighted the dramatic change this is in how we generate images! One question: have you found a great way to get similar results to "in the style of artists XYZ" without running into the content policy wall? This is one of the most frustrating parts right now for me (and I sort of get it why its in there). Otherwise the experience in ChatGPT + D3 is amazing. Also: give voice commands a try and use in different language -> its fantastic as an Image Gen Copilot for kids that way!
I actually haven't encountered the artist block that I was expecting. I prompt artists and get results. Do the results look like that artist? Sometimes. But that's the same with MIdjourney.
@@MakingPhoto defo less so than in midjourney. I feel like in MJ its usually way closer. Workound for me typially is to prompt to descreibe the artist style and then prompt the description. With mixed success though ;P
I’ve been having a lot of fun with this. Some things that are really difficult in Midjourney are easy here. For example, I find it almost impossible to get Midjourney to produce an image with three specific elements. Multi-prompt and image prompts help but not always. On the other hand I can’t get Dall-e 3/ChatGPT to make me clean sticker designs. I want them on a white background without a drop shadow. It will even say things like “here are four sticker designs without a drop shadow” yet every single one has a drop shadow 😀. Not that Midjourney doesn’t sneak them in, but at least sometimes it works. But definitely I’m loving this. LLM is the way forward
I'm having a similar experience. Some things like text are amazing in Dalle. Other thing s are better in MIdjourney. Try thinking laterally with ChatGPT. Ask for 2d, flat, line art stickers.
Just in the custom instructions how chatgpt responds you can say i want everything in 4k hdr or i dont want any digital or 3d render or i want different perspectives in each image. When i asked for never cgi it did so but i am still learning you could for example say put an elf in every image as a test or only this style or combine styles everytime.@@MakingPhoto
6:38 I don't believe DALL-E 3 has web access so what is likely happening there is ChatGPT is reading the URLs name or might be an image in the training set.
@@MakingPhoto Yes but web access is only under the Plugins or Web Browsing options from the GPT-4 menu. This also happens with default ChatGPT, it reads the URL text or has the data in its training data. You can talk to default GPT about Wikipedia URLs and it knows or guesses what it's about. Hopefully, OpenAI will merge all these into one big model. That would be awesome. I'm sure GPT-5 will be one big thing that does it all.
@@KolTregaskes thx mate... I was mindblown that this would have been possible, but upon trying I never got it to work. So I believe its what you are saying. Would have been a great way around the super annoying content policy restrictins for using artists styles :/. This is the part that bothers me the most, although I get it to a certain degree. Imho still one of MJs biggest USP with the artists styles and blending.
@@Blocky007 ChatGPT is a single modal model so the way OpenAI have done it is probably the only way they can do it. Though they are clever people so I'm sure they'll merge plugins, web, image, voice etc together at some point. If not with GPT-4 then I'm almost certain GPT-5 will be multi-modality and all these will be all-in-one.
Use the custom instructions in section in ChapGPT to instruct DOLL+E 3 to always create images in the 1:1 aspect ratio unless you tell it to use something else.
There are numerous issues with the DALL-E 3 service integrated with ChatGPT, including frequent bugs, interruptions, and system crashes. These problems persist even in the Beta version, making it evident that the system is not yet equipped to handle a large user base. What's puzzling is that these issues exist even for those who have subscribed to the Plus version. Charging for a service that's not fully functional is questionable. Despite these setbacks, the concept is fantastic. When it does work, the experience is excellent. I hope the service will improve soon.
I'm not seeing the same issues you are. The only thing I had was with an image not loading with a chat when I've gone back later. I'm not sure what these frequent bugs are - I've played with this a lot and am not seeing these. But with everyone using the system, especially at first, this doesn't surprise me. Bing is also getting slammed.
@@MakingPhoto I've already contacted the ChatGPT support team and followed their troubleshooting suggestions, such as switching browsers, changing networks, and opening an incognito window. Despite these efforts, the issues persist. The service functions well at times, but I've had two unproductive nights due to intermittent service outages. I'm now concerned that I might be the only one facing these problems. I'm quite frustrated and hope these issues are resolved soon. thanks, I love your channel.
Although I've been a huge fan of MJ and still think it has the best quality images, stuff like this makes me think it's going to get left behind REALLY soon 🤔
The future arrived earlier than I expected! If I remember correctly this chat model is not something that David Holz aspires to. I think if he really does not adapt he goes the way of the Blackberry.
?? As I said in the video - wide, tall, square are your aspect ratios. Don't expect EXACT aspect ratios. But you can get something other than square images. I don't know what you're telling ChatGPT - but it will zoom in and out -- see the video.
What are you having problems with? I did find it as easy as I show in the video. I don't tend to do complicated prompts. And if I have to try anything too many times, I get bored.
Wow! I'm absolutely mind-blown! Can't wait to try it!!!
Have fun!
Great, one of the best videos you have created. There were lots of "Oh, I didn't know that" moments in this video for me.
Great! Have fun exploring!
Stuff is so crazy. 😭 Been able to build a personal brand with just an android device and $0 startup. Won some giveaways and made some awesome connections. Can't imagine what is possible with a laptop, chatgpt and these new tools like avatar Creation, image animation and more. Completely mind blown with Dall e 3 quality.
There are a lot of tools out there, now! It's hard to keep up!
thank you, the video is fascinating and useful😊
You're welcome 😊
now we can refer an image. great.
I have been empowered by all of your amazing videos. I wonder if you know a way to create consistent looking characters in DALL-E 3 within ChatGPT. I wonder if there’s a way to use a particular image of a character and adjust it so that it looks the same in each image, but is changed sequentially from page to page. For example, a character for a children’s book that meanders through a story, and looks identical in every image, even though they are performing a different task throughout the story. In other words, it’s the same person throughout the book and you recognize that it is the same person or main character such as a cat or a dog. Cheers and thank you. Sincerely.
Thanks! I have explored this a little. Because chatgpt can remember, consistent characters should be easier. My results have been hit-and-miss. Sometimes I think ChatGPT has it figured out. Other times it doesn't. Let me play with this a little more. You should - in theory - be able to describe the character and ask ChatGPT to keep features the same. The character probably won't be EXACT, but close.
hi, thanks for the video, how can you specific the aspect ratio before the image is generated
Dall-e 3 isn't taking "specific" aspect ratios. You can ask for a wide or tall image (16:9) or a square image. Just ask when you prompt.
@@MakingPhotocould u plz write an example, say i want a 512x768, how can i write that in my prompt?
I don't think the tool is that precise right now. Ask for a tall image. "Make a tall image of a -----" You'll get something like 9:16. I did successfully get the AI to put a frame around the image for other aspect ratios, but it's hit-or-miss.
Wow. your video came just in time! Helped solved my problem with not being able to upload images under Dall-e3 mode.
Great to hear! I'll keep working on this topic. Let me know if you find any tricks!
ok that was amazing, I will have to get a plus account. Thank you for sharing your explorations, very inspiring!!
I've had a plus account as long as they've existed. ChatGPT is basically my PA these days. If I could plug it into my coffee maker, it would also bring me coffee in the morning.
my GPT-4 dalle can't access or view external URLs somehow.
After playing a bit more, I'm not sure the Dalle3 module is accessing the image. *sigh*
Amazing tips and way of presentation. I love that you highlighted the dramatic change this is in how we generate images! One question: have you found a great way to get similar results to "in the style of artists XYZ" without running into the content policy wall? This is one of the most frustrating parts right now for me (and I sort of get it why its in there). Otherwise the experience in ChatGPT + D3 is amazing. Also: give voice commands a try and use in different language -> its fantastic as an Image Gen Copilot for kids that way!
I actually haven't encountered the artist block that I was expecting. I prompt artists and get results. Do the results look like that artist? Sometimes. But that's the same with MIdjourney.
@@MakingPhoto defo less so than in midjourney. I feel like in MJ its usually way closer. Workound for me typially is to prompt to descreibe the artist style and then prompt the description. With mixed success though ;P
I’ve been having a lot of fun with this. Some things that are really difficult in Midjourney are easy here. For example, I find it almost impossible to get Midjourney to produce an image with three specific elements. Multi-prompt and image prompts help but not always.
On the other hand I can’t get Dall-e 3/ChatGPT to make me clean sticker designs. I want them on a white background without a drop shadow. It will even say things like “here are four sticker designs without a drop shadow” yet every single one has a drop shadow 😀. Not that Midjourney doesn’t sneak them in, but at least sometimes it works.
But definitely I’m loving this. LLM is the way forward
I'm having a similar experience. Some things like text are amazing in Dalle. Other thing s are better in MIdjourney. Try thinking laterally with ChatGPT. Ask for 2d, flat, line art stickers.
You can go to custom setting to fine tune chatgpt dalle-3 in a way of your choosing.
Tell me more of what you have in mind.
Just in the custom instructions how chatgpt responds you can say i want everything in 4k hdr or i dont want any digital or 3d render or i want different perspectives in each image. When i asked for never cgi it did so but i am still learning you could for example say put an elf in every image as a test or only this style or combine styles everytime.@@MakingPhoto
The elf test worked i said put in elf even if elf not in prompt then i typed a prompt sunset and it was sunset with elf.
Great tip!! I'll try this!
6:38 I don't believe DALL-E 3 has web access so what is likely happening there is ChatGPT is reading the URLs name or might be an image in the training set.
That's possible. But ChatGPT DOES have web access. So I'm not quite sure what it's doing. I'll keep exploring this feature.
@@MakingPhoto Yes but web access is only under the Plugins or Web Browsing options from the GPT-4 menu. This also happens with default ChatGPT, it reads the URL text or has the data in its training data. You can talk to default GPT about Wikipedia URLs and it knows or guesses what it's about.
Hopefully, OpenAI will merge all these into one big model. That would be awesome. I'm sure GPT-5 will be one big thing that does it all.
@@KolTregaskes thx mate... I was mindblown that this would have been possible, but upon trying I never got it to work. So I believe its what you are saying. Would have been a great way around the super annoying content policy restrictins for using artists styles :/. This is the part that bothers me the most, although I get it to a certain degree. Imho still one of MJs biggest USP with the artists styles and blending.
@@Blocky007 ChatGPT is a single modal model so the way OpenAI have done it is probably the only way they can do it. Though they are clever people so I'm sure they'll merge plugins, web, image, voice etc together at some point. If not with GPT-4 then I'm almost certain GPT-5 will be multi-modality and all these will be all-in-one.
Can someone please tell me how not to get Images that are cut off. I just want them to fit in the square.
Use the custom instructions in section in ChapGPT to instruct DOLL+E 3 to always create images in the 1:1 aspect ratio unless you tell it to use something else.
Dal-e 3 isn't in the chatgpt ios app yet.
It's in my IOS app. They are rolling this out. Not everyone's getting it at the same time. Keep checking your app.
There are numerous issues with the DALL-E 3 service integrated with ChatGPT, including frequent bugs, interruptions, and system crashes. These problems persist even in the Beta version, making it evident that the system is not yet equipped to handle a large user base. What's puzzling is that these issues exist even for those who have subscribed to the Plus version. Charging for a service that's not fully functional is questionable. Despite these setbacks, the concept is fantastic. When it does work, the experience is excellent. I hope the service will improve soon.
I'm not seeing the same issues you are. The only thing I had was with an image not loading with a chat when I've gone back later. I'm not sure what these frequent bugs are - I've played with this a lot and am not seeing these. But with everyone using the system, especially at first, this doesn't surprise me. Bing is also getting slammed.
@@MakingPhoto I've already contacted the ChatGPT support team and followed their troubleshooting suggestions, such as switching browsers, changing networks, and opening an incognito window. Despite these efforts, the issues persist. The service functions well at times, but I've had two unproductive nights due to intermittent service outages. I'm now concerned that I might be the only one facing these problems. I'm quite frustrated and hope these issues are resolved soon. thanks, I love your channel.
Dall-E 3 makes some really cute clip art for high school kids.
Sure - and some other things, too. If you know how to ask.
Coooool!
😀
GPT didn't want to go with a link as the style for an image. It just didn't do the trick for me.
Although I've been a huge fan of MJ and still think it has the best quality images, stuff like this makes me think it's going to get left behind REALLY soon 🤔
I understand completely! We may have some movement. The upscaler came out last night.
👋
The future arrived earlier than I expected! If I remember correctly this chat model is not something that David Holz aspires to. I think if he really does not adapt he goes the way of the Blackberry.
I admit, I like the ChatGPT prompting style a lot!
It doesnt worke. No aspect ratio, no Zoom in or out,
?? As I said in the video - wide, tall, square are your aspect ratios. Don't expect EXACT aspect ratios. But you can get something other than square images. I don't know what you're telling ChatGPT - but it will zoom in and out -- see the video.
@@MakingPhoto It's strange, I don't know why, on YT it works pretty well, but for me it makes a whole new design.
Chat GPT just put prompt engineers out of business!
Yes - probably. We'll see how this shakes out.
You are misleading people. It's not as easy as you make it seem. Not only is it not repeatable, but it also forgets rapidly.
What are you having problems with? I did find it as easy as I show in the video. I don't tend to do complicated prompts. And if I have to try anything too many times, I get bored.