Spaces:
Running
Tolkien's steampunk-style characters
poor golum 😄
I'm wondering what is the maximum resolution that can be squeezed out of this AI?
Dall-e mini has 256*256 dots. This can be easily viewed in the file properties.
However, I don't know the png format very well, maybe you can stretch it somehow without much loss
for example AI Midjourney, when you increase its resolution, it makes a picture that absolutely does not meet the requirements, but when you reduce the resolution, everything falls into place.
I haven't try the Midjourney, but as far as I can tell from what is written about these two AI, Dall-e has a more advanced algorithm.
We can hope that when the developers make the third version, the second one, with a resolution of 1024*1024, will be free available)
You can try Midjourney even now, there are no queues, you only need a discord, and they give you half an hour to try to work with it for free, though after the time has passed, either pay or go for a walk :) But she copes much better with faces and colors, you just need to try .
Yes, I know, thank you)
I just want to prepare more requests to use these half an hour as much as possible.
And while I'm getting ready, perhaps this or that AI will become public domain))
А ты по русски говоришь?
Yes, но как-то это нехорошо устраивать здесь междусобойчик)
как скажешь :)
By the way, some enterprising people are already made a website with their creativity based on dall-e mini
https://dallemini.com/
Do you know how to make it so that only the background changes during generation, the character remains alone and does not change?
No, I don't.
Is it even possible? As far as I know, in dall-e 2 you can edit the resulting image but here, probably, randomization has been added, which makes it impossible to get identical images for an identical request. As far as I know, maybe not.
жаль 😔
That is so unholy
hello :)) you know how to write a request correctly and write a refinement of this request in this request, but do it in such a way that the AI would accept it as one request. For example: "( face with beautiful eyes (eyes should be clear and super realistic)) and what would the AI take as a request for "face with beautiful eyes" so that it would not duplicate the eyes twice. Извини если слишком замудрил.
I'm afraid not
This will already be a kind of recursion.
It is necessary to understand how the GPT-3 language works.
As far as I can see, there is a principle of relevance, that is, certain words have more meaning than others.
For example, a hybrid of Gollum and Chekhov will always be more of a Gollum, because its relevance is higher.
Here, there is such an article, interesting but longread:
https://openai.com/blog/dall-e/
for some reason, this AI has problems with photographs, it still copes with drawings somehow
It is difficult to take good photos with a resolution 256 * 256 pixels)
And dall-e was originally conceived as a tool for working with images and art, not photos. Probably, the neural network was taught more drawings by different authors than photographs.
A great portrait, I wouldn't improve anything