You joke about being an expert in this sort of image generation, but you must be in like the top 0.001%. Maybe you're not THE world expert, but as a person who does this every single day, you're considerably more knowledgeable than I am, at least in terms of what's going on under the hood. And, you've done a great job of keeping up with what the models can do - I am often surprised to hear about things w/a model I've been using for a while!
Are you finding yourself using much Sora? It seems to have fallen way to the back burner of my interest lately, but is it worth a revisit now?
If you mean Sora the video model, then no. It's definitely not top-tier at this point, and you have many better options like Kling, Veo 2, Hailuo, Wan 2.1, etc.
But Sora the platform where you can use presets as I've shown, sure thing - it's a handy way to make GPT-4o images. Since you're already making GPT-4o images in ChatGPT, you should give Sora a try for sure.
When i go to Sora.com you link, it wants me to open the ChatGTP app and when i do well, no sora. So I’m confused (also lazy because I got my coffee and pink blankie and giving up the ipad for a pc seems like too much work rn)
I finally coughed up my $20 for the image generation. It’s amazing. I used it to design a sticker and I noticed something new. First, it worked a lot like your logo prompter by translating my freeform text into bulleted list of requirements for me to review and refine. Then, when I said go, it created a prototype (!) for my approval. Only after that did it generate a high res proof. I popped that into canva and made 1 edit (need a QR code) and PDF right to the sticker makers. Have you seen that prototype workflow? Also, my sticker was based on Spider-Van and at first it said it couldn’t do it but just saying make it Spider-Man *like* was enough to keep going
I'll find the text later and share. I also have to share my sticker. After a few iterations it was still kinda flat so I asked it to give me some ideas for action and it did and rendered them staying petty true to the original design. Then I asked it for ideas to add another color and make it more comic-booky and it did that too. The combo of back and forth conversation based refinement and then it nailing it in the image is the THING.
Thank you for a short explicit demo on several tips. Although I have been incorporating these ideas of combo prompting and reference images for awhile, I got a couple new tips out of it which is always beneficial and saves me the drudge of playing with it to figure out the best wording.
Thank you again for the demo, Daniel! Here at Futuristic Lawyer FM, we are looking forward to welcoming you back again on the station as AI's image generation capabilities progress
I need to try using the reference images for colour palette and style, they look so good!
I recently learned that Van Gogh was massively influenced by Japanese art, and even did some copies of Hiroshige paintings, so I created a preset using those and it's so good. I need to try to make some more. It's very under utilised.
You joke about being an expert in this sort of image generation, but you must be in like the top 0.001%. Maybe you're not THE world expert, but as a person who does this every single day, you're considerably more knowledgeable than I am, at least in terms of what's going on under the hood. And, you've done a great job of keeping up with what the models can do - I am often surprised to hear about things w/a model I've been using for a while!
Are you finding yourself using much Sora? It seems to have fallen way to the back burner of my interest lately, but is it worth a revisit now?
I do try to keep tabs on things, but it's a LOT!
If you mean Sora the video model, then no. It's definitely not top-tier at this point, and you have many better options like Kling, Veo 2, Hailuo, Wan 2.1, etc.
But Sora the platform where you can use presets as I've shown, sure thing - it's a handy way to make GPT-4o images. Since you're already making GPT-4o images in ChatGPT, you should give Sora a try for sure.
When i go to Sora.com you link, it wants me to open the ChatGTP app and when i do well, no sora. So I’m confused (also lazy because I got my coffee and pink blankie and giving up the ipad for a pc seems like too much work rn)
Do you have ChatGPT Plus? Sora was initially for paid accounts only, so perhaps that's why they want you to link the two?
I do now, is Sora the same ChatGPT UX or diff?
Sora is a completely different standalone platform and interface. Less chat, more prompt field and visual tiles.
I finally coughed up my $20 for the image generation. It’s amazing. I used it to design a sticker and I noticed something new. First, it worked a lot like your logo prompter by translating my freeform text into bulleted list of requirements for me to review and refine. Then, when I said go, it created a prototype (!) for my approval. Only after that did it generate a high res proof. I popped that into canva and made 1 edit (need a QR code) and PDF right to the sticker makers. Have you seen that prototype workflow? Also, my sticker was based on Spider-Van and at first it said it couldn’t do it but just saying make it Spider-Man *like* was enough to keep going
Oh wow, that's awesome. Don't believe I've seen it propose images as rough drafts before, only text descriptions.
And yes, the ‘I don't want Mickey Mouse, I want a mouse that looks like him” workaround is a thing. Not always flawless but it's worth a try.
I'll find the text later and share. I also have to share my sticker. After a few iterations it was still kinda flat so I asked it to give me some ideas for action and it did and rendered them staying petty true to the original design. Then I asked it for ideas to add another color and make it more comic-booky and it did that too. The combo of back and forth conversation based refinement and then it nailing it in the image is the THING.
a Sora option showed up in custom GPTs area today :) I think the Library showed up this week too although maybe that's because I just upgraded to plus
Here's the walkthrough text for the mockup process:
🎨 Next:
I can sketch a rough mockup showing the layout right now if you like, to make sure you love it before going further.
After that, if you want a final digital version ready for print (like SVG or PNG), I can prepare that too!
Question:
Would you like a quick rough sketch mockup first? (I'll make it clear and simple so you can approve or tweak anything before the polished version.)
Thank you for a short explicit demo on several tips. Although I have been incorporating these ideas of combo prompting and reference images for awhile, I got a couple new tips out of it which is always beneficial and saves me the drudge of playing with it to figure out the best wording.
Awesome to hear this is useful to you, Nicole! What kind of styles do you usually explore in GPT-4o?
Thank you again for the demo, Daniel! Here at Futuristic Lawyer FM, we are looking forward to welcoming you back again on the station as AI's image generation capabilities progress
It was my pleasure. Thanks for having me, Tobias!
I'd love to join again in the future to explore---and have an existential crisis over---new AI features.
Haha great
I need to try using the reference images for colour palette and style, they look so good!
I recently learned that Van Gogh was massively influenced by Japanese art, and even did some copies of Hiroshige paintings, so I created a preset using those and it's so good. I need to try to make some more. It's very under utilised.
Awesome to hear you're already using these options! Feel free to share some of the most impressive images you end up creating!