AI Imagelab: Everything about AI generated images and videos

I prefer Leonardo.AI (a stable diffusion variant) due to its faster processing, better options (privacy, finetuned models, less censorship in prompts) and pricing. Midjourney is superior in image quality by my tests, but not by a huge margin. Best regards Mattes

Midjourney introduces two new features that make it easier to generate additional variations of an image: Repeat and Permutations I'll explain how they work in a moment - but here's the catch: These features are only available for the most expensive paid version (PRO for $60 per month), as they consume huge amounts of processor power. Introducing the Repeat feature Type --repeat with a number are your job to run that job that many times. For example /imagine cats --repeat 5 will create five 2x2 grids of cats Try mixing this with other arguments like --chaos to radically increase your pace of exploration Introducing the Permutations Feature With this function you can define a large number of jobs with different word combinations in one go- A simple example of permutations: /imagine a {cyberpunk, vaporwave, art deco} (cat, dog) This will generate all the following jobs /imagine a cyberpunk cat /imagine a vaporwave cat /imagine a art deco cat /imagine a cyberpunk dog /imagine a vaporwave dog /imagine a art deco dog You can even nest {} inside other {} - but be careful, this can quickly escalate into extremely many jobs.... /imagine A {sculpture, painting} of a {seagull {on a pier, on a beach}, poodle {on a sofa, in a truck}}.

Here's an example: From the image of a Chinese park backdrop (a real photo, really), Midjourney generated four different descriptions, one of which in turn was used as a prompt for the new images:

Since recently Midjourney has a new command that reverses the previous way: Using /describe, text-to-image becomes image-to-text - Midjourney provides a description to uploaded images. Tonight, the function was announced via Twitter: The feature is still so new that it is not listed in the Midjourney command documentation (as of April 4, 2023). Moreover, it still seems to be in beta, since the command currently generates error messages reliably. On its own, this has limited utility, but it does open up new possibilities. Prompt templates for new images can be generated from existing images Alternative texts can be generated for images on the web, which is an enormous relief for people with visual impairments

Camera Perspective, Cropping and Alignment in Midjourney V5 Midjourney has been incredibly good at generating photorealistic images of people since version 5 at the latest. After the first experiments, however, the desire to influence the result sets in: Besides the subject itself and the "look", classic parameters from film and photography are a real challenge, especially perspective and cropping. How do I get a close-up from the frog's perspective? Or that the whole body of a person is shown from above? And that the model looks to the side? In this tutorial we will give an overview of different camera perspectives and image details and how to persuade Midjourney to generate appropriate images. Camera Perspective Let's start with camera perspectives. As a basic  prompt Midjourney gets the task to show an old cowboy. We always use the same seed, so that the results keep a similar style and use different instructions for perspectives and then later sections. /imagine [PHRASE] of an old cowboy, shot on Kodachrome, natural lighting --seed 1 Eye-Level Unless otherwise specified, Midjourney generates images from eye level, so this perspective usually does not need to be explicitly specified. /imagine Eye-Level of an old cowboy, shot on Kodachrome, natural lighting --seed 1 Low-Angle Important is the phrase low-angle, if the effect is not sufficient, you can add from below . /imagine low-angle photo of an old cowboy, shot on Kodachrome, natural lighting --seed 1 Extreme Low-Angle By adding extreme you can reinforce all specifications, so you get a frog perspective, for example: /imagine extreme low-angle photo from below of an old cowboy, shot on Kodachrome, natural lighting --seed 1 High-Angle The phrase high-angle does not always work, if necessary experiment with the addition from above especially when working with cropping. /imagine high-angle photo of an old cowboy, shot on Kodachrome, natural lighting --seed 1 As alternative option: /imagine photo from above of an old cowboy, shot on Kodachrome, natural lighting --seed 1 Extreme High-Angle If you want pictures from a much higher point of view, you have to add the phrase extreme here as well: /imagine extreme high-angle photo from above of an old cowboy, shot on Kodachrome, natural lighting --seed 1 Birds-Eye-View To generate images with a perspective vertically from above, there are various phrases you can experiment with. I have not been able to get a really vertical perspective from above for people pictures. You can test: /imagine Birds-Eye-View photo from above of an old cowboy, shot on Kodachrome, natural lighting --seed 1 /imagine drone view photo from above of an old cowboy, shot on Kodachrome, natural lighting --seed 1 /imagine Drone Shot Photo of an old cowboy, shot on Kodachrome, natural lighting --seed 1 /imagine Aerial photo of an old cowboy, shot on Kodachrome, natural lighting --seed 1 Cropping In the next step, let's look at the different image details. The names and corresponding phrases are familiar from photography and film. We use the same basic prompt again, but now add instructions for the section: /imagine [PHRASE] of an old cowboy, shot on Kodachrome, natural lighting --seed 1 Extreme Closeup With the addition of extreme close-up, you get portraits where the face fills the frame completely. /imagine extreme close-up of an old cowboy, shot on Kodachrome, natural lighting --seed 1 Closeup Phrase close-up /imagine close-up of an old cowboy, shot on Kodachrome, natural lighting --seed 1 Medium-Full Shot /imagine Medium-Full Shot of an old cowboy, shot on Kodachrome, natural lighting --seed 1 You can also use Medium-Long to achieve similar results. Medium-full or long shot prompts usually show more of the subject's body than a Medium Shot. Full-Body Shot The phrase Full-Body provides for full-body shots, but the feet are often cut off. Other aspect ratios do not help here. /imagine full-body photo of an old cowboy, shot on Kodachrome, natural lighting --seed 1 Positioning As a third category we can still position the subject in front of the virtual camera: Front View  Default is front view, I got good results with front view . /imagine Front view of an old cowboy, shot on Kodachrome, natural lighting --seed 1 Side View With side view you get a 90° side view of the subject. /imagine side view of an old cowboy, shot on Kodachrome, natural lighting --seed 1 Back view The command from behind provides a back view. /imagine from behind shot of an old cowboy, shot on Kodachrome, natural lighting --seed 1 Inspiration: https://twitter.com/nickfloats/status/1640404667036483586