DALLE3 Rolling out to ChatGPT Plus members

There was no credit system on ChatGPT Plus DALLE3, but there were limits on number you could generate per hour, I believe…

Oh, I see… Thank you for your support!

I dislike this limited rollout system. I pay for ChatGPT so that I can use the new features. Why do I have to wait up to a month just to try them, when it’s available for free on bing?


Welcome to the forum.

Yeah, it can be frustrating. However, I think it’s due to being cautious and keeping safety in mind - as well as ensuring the servers don’t melt with all the traffic. The better quality is worth the wait, imho.

Thanks for the welcome!

I get where you’re coming from and there is several points that could be brought up against what you’ve said, like free-use bringing more users than the paid ChatGPT+ but, there is no point crying over split milk.
The video demo of Dalle-3 almost shows “editing” in a way, when you ask a follow up, or for a change.
As someone with access does it seem like something different than just remaking the prompt to include your changes?

No problem. I do see it from your side too.

Basically, the ChatGPT Plus implementation seems to be a new way to work with DALLE3 in that you ask for what you want, but you don’t need a super-technical “prompt” like MidJourney, etc. - just natural language. Your request is then formed into prompts by (GPT-4?) to get the resulting images. This allows you to “edit” in a sense until you get what you want. That said, at this point, you can still force it to produce your exact prompt, but that will be available via Labs and API too (I think…)

So, yeah, a new way to interact with the model.

If you give me a prompt or idea, I can run it and show you what I mean in action.

Well I guess an interesting test would be to do something similar to the video.
Ask for an orange cat named “Yuzu” and then, in the follow up, ask for the same cat but with a different background, and just see how much the original “Yuzu” changes. That’ll be clear if there’s some trickery going on past just prompts.

But feel free to just do some random prompt with a cat, to show what you mean!

There’s no way to “download all images” in a thread or anything (yet?), but I’ve heard whisperings of people working on tools to make it easier to manage. I think for mass production, though, Labs or the API will be better when they’re available for DALLE3.

You can control it a bit with ChatGPT - and as you see on the last image - even take a generated one and make it landscape instead of square… (Notice it said “first image” but accurately grabbed the second image as requested… so it’s still a bit wonky in some ways, but still pretty amazing…)

1 Like

I see! So it does have some cool tools! That’s outpainting you’ve shown here! I was worried it was just going to be the bing image generator in ChatGPT but that doesn’t seem to be the case. Now I’m even more excited to see what it can do.

EDIT: Oh wait, it isn’t outpainting… it’s just very similar… how the hell did it do that?
My best guess would be that ChatGPT was able to see the seed used to generate the initial image, and reused the same seed but in landscape

1 Like

Right?! :wink:

… the leading theory is that there’s a “seed” so exact same prompt will give exact same image. No word if this will always be the case, but I’m thinking it will be for ChatGPT Plus DALLE3 implementation?

Myself and others in the alpha group have asked for the ability to turn the seed on and off, but no word on if that will be available in ChatGPT Plus, Labs, or API… Still, it’s kinda cool and helps if you want to try to get shots of the same type of model…

I’ve run a lot more tests on Twitter…



I tried this too… it’s not the same pose, but one could argue the cat and style are the same?

I had already made orange cats before I saw your post. With remarkable results.

The AI was able to make my 3D object and envision it from different perspectives without going haywire.

And as long as we are going for IP infringement…

The computer-vision feature is great, imagine if you could combine them to get controlnet or img2img like results with Dalle-3.

1 Like

ControlNet with Dall-e would be epic. I can only wish, but I’m not hopeful. But if it did I would almost instantly start working on some consistent video-making :drooling_face: like WarpFusion

I explored this a bit for someone (scroll up from the link to see the source description). Image prompt and vision description are a far distance from meeting as far as language goes.

I eagerly await Dalle-3 and the multimodal features its actually why I purchased plus. A point to chatgbts dalle 3 vrs bings image create is with bing you have no commercial rights to the images you create. from what I can tell with gbt you may use your images for what ever you please

1 Like

Will Plus members be notified when it becomes available? I asked Chat if it can generate images now and it said yes. Got me pretty excited. but then i would only get weird prompts back.

You have to choose the DALLE3 model. I’m not sure if it’s rolled into the other models yet. Upthread, there’s a screenshot of what it should look like. You might need to refresh.

Yeah, i tried refreshing… still no good…

still just get this…

also attached is what it says when i ask it if it can.