I’ve spent 1000+ hours working with Ai image generators, Flux is the best by a long shot. I use Getimg to access it and test, you can choose how you want to get there.
TL:DR here’s the conclusion. To get the highest quality/consistency out of Flux, eliminate as many variables as possible. I’ll show you my boilerplate prompt I’ve been using where there are only four main variables, you can obviously edit for your needs.
Image generation
Here’s the generic prompt, I’ve highlighted the variables.
Create a Pop art style of a person sitting on a couch. Centered vertically and horizontally taking up 1/3 of the space of the overall image. make the background a well appointed home.
Variable 1: Style. You can see a list of styles here but there are so many you can just ask your favorite chatbot for a recommendation.
Variable 2: The subject. I found that having a single subject of the image increases quality/consistency dramatically. Having Flux focus on one thing at a time helps a tremendous amount.
Variable 3: Positioning. Without positioning it gets frustrating because the subject keeps moving around and the aspect ratios are all off. Define it exactly. Still isn’t always consistent but you can get there.
Variable 4: Background. Simplest is a color or gradient, most complicated can go on for a while. The more complicated the more likely it is to break in my testing.
Let’s look at some examples.
Create a Van Gogh Style of a cat sitting by a window. Centered vertically and horizontally taking up 1/3 of the space of the overall image. make the background purple hues.
Create a minimalist style of a graphic of a home interior. Centered vertically and horizontally taking up 1/3 of the space of the overall image. make the background yellow gradient.
Create a photorealistic style of a person on a laptop. Centered vertically and horizontally taking up 1/3 of the space of the overall image. make the background red.
Create a brutalist style of a gothic house in a storm. Centered vertically and horizontally taking up 1/3 of the space of the overall image. make the background grey skies.
create a high res landscape photo of half dome. centered vertically and horizontally on a 9:4 aspect ratio background make the background gradient purple
create a high res photo of a mid century modern living room. centered vertically and horizontally on a 9:4 aspect ratio
Video generation
The image to video is not bad honestly.
Original prompt turned to video
create high res landscape photo of five balloons rising over the Turkish countryside. centered vertically and horizontally on a 9:4 aspect ratio background. Angle should be looking up from the ground. make the background a gradient colored sunset
create a product photo of a metal re-usable water bottle with a popped up lid. centered vertically and horizontally on a 9:4 aspect ratio. Background is purple with shading for a light source coming from the top right.
create high res landscape photo of half dome with clouds in the background. centered vertically and horizontally on a 9:4 aspect ratio background make the background gradient purple
You can play around with each of the variables to test the limits. I typically like to see how far you can push the limits before it breaks. Then see if there’s a way to fix whatever is breaking.
Known limits.
- A person doing an action with another object in the picture. Like a person on an exercise bike, the interaction between the person and the object is still very inconsistent.