Are you struggling to get good images out of DALLE 3?
Jakob Radb?ck
F?rel?sare & poddare inom AI & e-handel | E-handelskonsult | Tidigare e-handelschef p? MediaMarkt | Utbildar f?retag inom AI-effektivitet och arbetsgl?dje
Hey everyone!
Today I will talk about DALLE 3, an AI image generation tool by Open AI.
Often DALLE 3 gets overshadowed by the popular tool, Midjourney.
But wouldn't be neat to have everything under 1 roof (and 1 subscription for that matter..)?
As I have previously mentioned, in generative AI, there is actually something called "stupid questions". Bad questions (prompts) will give you bad results, period.
The skill you have to master in order to get great results is effectively structuring your prompt.
In below example we will focus on creating a portrait image, but the structure can be used for any type of image.
Here how:
Start with a brief description of the image you want to create.
#Top level description of your image
Now, create 1 section for each part of the image, something like this.
#Scene: (Here you give all the information of the scene of the image)
#Person details: (Here we give all information of what the person looks like, is wearing, body composition, hair, skin tone, face expression, eyes of color etc.)
#Lighting: ( For example: Natural light, global illumination, backlight etc.)
#Settings: (Here you place all of your specifics like camera, lens, style, etc)
The end result would look something like this:
Create an image using this exact prompt:
领英推荐
#Description
A hyperdetailed, photorealistic black and white portrait of an elderly man, exuding a blend of wisdom, resilience, and a life rich with stories.
#Scene
The scene is set against a completely black background, offering a profound contrast that accentuates the man's rugged features. This minimalistic backdrop serves to draw the viewer's attention solely to the subject.
#Person Details
The old man has a weathered yet dignified face, marked by deep lines and creases telling of his age and experiences. His beard is thick and unkempt, streaked with shades of grey and white, resembling the bristles of a seasoned warrior. Intricate face tattoos, each with its own history, adorn his arms and visible parts of his chest. His eyes, almost black, are deep pools of knowledge and mystery, reflecting a lifetime of wisdom and untold stories. There's a subtle, almost imperceptible smile playing on his lips, hinting at a quiet contentment with life.
#Lighting
The lighting is a masterpiece of natural light, combined with global illumination and uplight f/1.8, casting a gentle yet revealing glow on the man's face. It creates a play of light and shadow, highlighting the textures of his skin, the roughness of his beard, and the intricate details of his tattoos.
#Settings
The image is captured in a cinematic 16:9 aspect ratio, resembling a frame from a high-definition documentary. It's shot using a Canon EF 16-35mm f/2.8L III USM lens on a Canon EOS 5D Mark IV camera, ensuring unparalleled clarity and depth. This setup is not just about capturing an image; it's about immortalizing a moment in time, telling a story through the lens with utmost precision and artistry.
Let's take this bad boy and copy it over to DALLE 3 to generate our image.
??.......Here's the result!
TIP: Sometimes DALLE 3 is doing some pre processing to your prompts. When doing advanced prompting you want to avoid this. Avoid this simply by writing "Create an image using this exact prompt:" before your prompt. This tells DALLE to use your exact prompt without pre processing.
Thanks for reading all the way down here ??, I'd love to see your creations in the comments!
Have a really fantastic day and rest of the week!
Stay curious ??
#DALLE3 #chatgpt #OpenAI #AIart #GPT4
Senior Software Engineer at eGrowcery
10 个月For me, any mention of a camera type or photograph/photography when using Dall-E 3 through the API seems to want to quite often put a camera in the image, or a camera taking the picture in the image. Additionally, the Dall-E 3 engine often puts textual gibberish and lines on the image when it doesn't seem to understand some of the prompt. This doesn't happen with the older Dall-E 2 engine. Any hints appreciated. eg trying this prompt (with and without the "using this exact prompt": Create an image using this exact prompt: #Description A hyperdetailed, professional food photography of the freshly made recipe "Oatmeal Cakes". #Food Details A summary of the food is "Delicious oatmeal cakes that are perfect for a hearty breakfast or a nutritious snack. These cakes are easy to make and are a great way to start your day.". #Scene The scene is the food served on a table, with the food in sharp focus and detail. #Lighting The lighting is perfect nautral lighting, emphasizing the textures and colors of the food. #Settings The image is captured in high quality, resembling a photo for a food magazine. It's shot using a Canon EF 16-35mm f/2.8L III USM lens on a Canon EOS 5D Mark IV camera, ensuring unparalleled clarity and depth.
Physics at UC Berkeley
1 年This is gonna come in handy for me, thanks!
Content Producer & Social Media Specialist at Fagerhult. With sustainability and smart lighting at heart, we create light for better living.
1 年VERY GOOD good guide for a beginner like me. Haven't started prompting yet but this gave me a better understanding of the logic. Will save this for later!