If you use a lot of chatgpt, you have probably experienced with Dall-E 3, the latest version of an OpenAi AI image generator. Dall-e 3 can be surprisingly good to give life to his ideas, but sometimes surprisingly bad to understand certain details or simply surprising in what he chooses to emphasize. Even so, if you want to see that a specific scene moves from your head to your screen, Dall-e 3 is usually quite useful, you can even make your hands write.
But here is the thing, Dall-e 3 is still a model of AI, not a mental reader. If you want images that really look like what you are imagining, you must learn to speak your language.
After a little trial and error (and some involuntary horrors), I have become quite an expert in speaking their language. So here there are five key tips that will help you do the same.
Polished in HD
Dall-e 3 has a quality mode called ‘HD’, which makes the images look more clear and more detailed. Think about it how to add a high -end camera lens to your art generated by AI: textures are richer, lighting is more refined and things like fabric, skin and reflexes have more depth.
See how it looks in this notice: “A representation of a close -up photo of a butterfly that rests on a sunflower. Quality: HD”.
Play in appearance
Not all images have to be a square. Dall-e 3 allows you to establish aspect relationships, which may seem less, but it may be huge if you want to make something more cinematographic or as a portrait.
The square is fine, but why limit yourself when you can do something epic? This is especially useful for the graphics and posters of social networks, such as the next one, which had the notice: “A vertical poster of an ancient ads for Paris, size 1024×1792 pixels.”
Think like a film director
To get an image capable of evoking emotions, sometimes it helps to think as if you were a photographer in the real world. Think about the camera angle or composition techniques; Look for them if necessary. The result can drastically change how an image is seen.
Instead of a flat and dead image, you can request angles such as foreground, bird view or excess shoulder. The same goes for composition styles and terms as ‘symmetric composition’ or ‘field depth’.
This is how you can obtain the following image of this notice: “A dramatic shoulders of a lonely cowboy standing in a rocky cliff, looking at the vast desert landscape underneath. The sun is put in the distance, throwing the canyon long shadows.
Iterar, Iterar, Iterar
One of the less known but highly effective tricks of Dall-e 3 that is to tell you what to not include. This helps avoid unwanted elements in your image. That could mean specifying negative elements such as colors, objects or styles that you do not want or refine the style and mood so you do not want it to feel.
This is how I obtained the image below, using the notice: “A quiet park in autumn with a young woman sitting in a wooden bank, reading a book. The golden leaves cover the ground and a soft breeze whispers the trees. None other People, without a litter, just a quiet and serene moment in nature. “
Be too specific
Think of Dall-e 3 as a very literal genius: it gives exactly what you ask, no more, nothing less. Then, if he writes “a dog”, do not be surprised when a random dog, undetermined moral alignment, environment or alignment. The more details include, such as race, color, environment, mood or even art style, the better the results will be.
As an example, you can start with: “A wizard that throws a spell”, but it would be better for him to send: “An elderly wizard with a long and braided white beard, dressed in emerald green robes embroidered with gold runes, conjuring, conjuring , conjuring a whirlwind vortex with blue rays of their fingers in a stormy mountain landscape. ” You can see both below.
