Canvas Whisperer: Hyperrealistic Artwork

[zealous nacelle]

Canvas Whisperer elevates your visions into stunningly hyperrealistic art pieces. Be it a peaceful landscape, a bustling urban scene, or a lifelike portrait, your ideas are meticulously transformed into rich, detailed imagery, offering a nuanced visual experience.

Ideal for art aficionados, interior designers, and those seeking to infuse their spaces with unique, bespoke elegance.
https://chat.openai.com/g/g-wGnVJgUWU-canvas-whisperer-hyperrealistic-artwork

[frank kelp]

bro

🤯

[grand pike]

i love that desk tbh

[zealous nacelle]

Updated for better result output.

[hybrid crypt]

Really nice

[zealous nacelle]

Much appreciated 😀

[lusty dragon]

wow

[spice flower]

Precoux and Eskcanta approve!

[frank kelp]

@zealous nacelle where we started

where we got to

[latent thorn]

Brilliant thanks

@zealous nacelle what kind of credits do you want for this?

[zealous nacelle]

Not sure what you mean?

[latent thorn]

For any images I create using this, do you want some kind of credit?

[zealous nacelle]

No credits needed. Any image you create is yours.

[latent thorn]

I appreciate it

[frank kelp]

Is this Just Using Dall-E or anything else under the hood?

[zealous nacelle]

It's Dall-E and a specific set of guidelines I made using my other GPT: https://discord.com/channels/974519864045756446/1172301858300235846

[frank kelp]

what i'm trying to figure out is How APIs are used within the GPTs like there's this dad jokes one as well

[spice flower]

Glorious. I feel like I'm there!

Oi. ran into this. Checking with you in case it's more likely to be your instructions instead of the base model's issue, I don't know.

[zealous nacelle]

That's from me. I'm constantly updating it
Is is fitting to your liking or is it deviating?

[spice flower]

My worry was, it didn't make the image

I had to ask twice

[zealous nacelle]

Probably because I was editing the instructions. Try again and let me know.

[spice flower]

I like your style. I'm now advertising you sometimes, where seems appropriate. #prompt-engineering message

[zealous nacelle]

Thank you so much! 🙏
I really appreciate that

[jade dust]

good morning

[zealous nacelle]

Updated.

[latent thorn]

truly an amazing gpt

[pure shale]

We'll see🧙‍♂️

[zealous nacelle]

Thank you! ❤️

[thorn obsidian]

I attempted to generate a random image based on your request for a random creation, but unfortunately, it seems the request didn't align with the content policy guidelines. As a result, no images were produced.

[zealous nacelle]

That policy is universal to all image generations using Dall-E and ChatGPT, following copyright laws. Ask it to create something similar, from an alternate reality or something along those lines.

[spice flower]

Users are going to hope and trust that you have instructions inside your GPT that will guide it to follow especially your recommended requests to get a picture.

[thorn obsidian]

I picked random, it couldnt do it, tried again and got the same issue, will try later

[spice flower]

While the problem is universal, yes....

Could you in your instructions guide the model to 'do it right'?

That's why we come to GPTs, at least why I do.

I think you know how to talk to the AI for this purpose, better than I do (And I'm Esk.... I think I'm good at prompt engineering in general 😄 )

And I expect you to have set it up, Especially if I decide to use your preset recommended question, to get a successful picture.

Also, if I click your preset, which I presume is what GeGeek is doing?

I can't add extra instructions, the AI goes with that exact wording. So it's 100% in your power, Py, if we succeed or fail with your preset recommended questions.

Therefore, try to ensure you recommend only the best that can be gotten using your GPT.

Is my recommendation.

[zealous nacelle]

I'm not sure if I would feel comfortable trying to find ways to bypass OpenAI's content policy by forcing the likeness of copywritten artwork. OpenAI has those policies in place for a very specific reason.

And for the presets, I believe those are just examples, and not meant to be used to add extra instructions. However I will change them from both of your requests to optimize a better result.

[zealous nacelle]

New edits complete. Let me know what you guys think. ❤️

[spice flower]

oh gosh. I am so sorry. I do not mean I think you should work around OAI's safeguards, that is not what I meant.

I meant, I think it's reasonable for you to instruct your model itself to avoid likely copyright and other issues, especially when your custom built 'ask it this' questions are used, because if a user clicks those, we can't add extra stuff.

Where you told GeGeek, "Ask it to create something similar, from an alternate reality or something along those lines."

He can't, if he's clicking on the buttons you built into your GPT.

You need to have that in the instructions for us?

And since that's 'make something random', and a built in recommended question, you can also guide your GPT to 'randomly' pick stuff that isn't likely to get rejected for content. GeGeek claims they used it 2x and got failed gens 2x. That may not be rare, and is likely fixable inside your instruction.s

[zealous nacelle]

I will do my best in trying to find ways for the AI to avoid copywritten material and to generate images in a rudimentary way, without bypassing content policy. Although, I have tried those types of attempts in the past with custom instructions and the AI either forgets, ignores, or figures out a way to make my specific command seem unimportant. But nonetheless, I will see what I can do.

[spice flower]

Aha. If you want to discuss the prompt engineering side of it... I love that stuff. Here, or #prompt-engineering, or wherever, if you do want help. It may be possible - if the AI can do it with direct instruction, it is possible...

If it isn't doing it in the complex GPT you made, that means, there is very likely some conflict in instructions somewhere.

Now, if you ask the AI to do something as a single direct instruction, and the AI doesn't do it well, that strongly suggests that the AI cannot do it well and there may not be a fix until a new model is released.

But again, if it can do it 'as a stand alone instruction' then it can do it inside a GPT.

But a GPT is often a lot of instructions. Even 1 conflict and you are likely to get guesses, and guesses might perform against expected or intended behavior.

I am happy to help you look for conflict, if desired, but that's what I'd be looking for.

[zealous nacelle]

Well, that's going to be an issue, because I am capped to the brim with 8000 characters for its complex instructions

I do use contradictory language though, for its instructions. But that's part of my intention, for optimization purposes. So that may be the problem. But I think it's more so that the AI cannot detect the intricacies of my contradictions and there may not be a fix until a new model is released.

[spice flower]

Yep. Some stuff can be reworded for efficiency - or you can just let it go. This is your model, your baby. If you're happy with it sometimes failing to gen images that you directly recommend users ask for - ...

[zealous nacelle]

I'm putting my faith in the OpenAI's hands to come up with better updates. I've done all I can do for this specific GPT.

[spice flower]

You're still the author of my favorite image-stylist GPT. I've compared a few others, and this one, to my tastes, is the best match so far.

And I am not coming to you for random images, so I personally am not affected by that flaw. What I am seeking, I am finding here.

[zealous nacelle]

Thank you so much. That means a lot. I'm so very happy to hear that

[zealous nacelle]

I encourage people to use the companion GPT #1172993830241443940 to enhance the quality of your images.

[frank kelp]

this continues to blow my mind but also saddens me as a landscape photographer