Stability AI
Featured images with Stable Diffusion 3.5 Large
The Stability image model is free for all users while it is in beta - so give it a try and give us some feedback
Overview
Machined can generate featured images using Stability AI - we currently use the Stable Diffusion 3.5 Large model.
You can configure automatic image generation for articles in clusters or manually create images for each article.
Machined provides 70+ pre-defined styles and let's you define the hue and tone of your images.
Usage
Automatic
You can enable automatic Stability image generation for featured images, simply select Stability in the Featured Images dropdown on your cluster, then select the settings you wish to use.
Each time an article is written in this cluster, a featured image will be created with Stable Diffusion using the settings you have selected.
Manual
You can also generate featured images manually using Stability, open the featured image popup for the article in question and navigate the the Stability tab.
From here you can generate images in two ways:
Using the Style and Hue approach
Define your very own custom prompt
Storage
Machined does not remove existing images when you generate new ones, so you can always go back to a previously generated image. You can see all images for an article by opening the Featured Images popup as described above.
Machined will store images for up to 3 months, at which point they will be permanently deleted, so please make sure you export or download all the images that you want to keep.
Pricing
Using Stability to create images is free while it is in beta. Long term pricing will be announced soon but will likely require Replicate api keys
Settings
Machined allows you to use styles and hues when generating images, to help you keep consistency between images in your clusters.
Image size
Machined generates images that are 1366x768 in size
Styles
The image style lets you define which artistic style the image will be created in. Some styles are more suitable that others depending on your niche, topic and keyword.
Below are examples of the available styles, generated for the keyword "how long do swans live".
Abstract Art
Abstract Geometry
Art Deco
Art Nouveau
Bauhaus
Bokeh Art
Brutalism in design
Byzantine Art
Celtic Art
Charcoal
Chinese Brush Painting
Chiptune Visuals
Concept Art
Constructivism
Cyber Folk
Cybernetic Art
Cyberpunk
Dadaism
Data Art
Digital Collage
Digital Cubism
Digital Impressionism
Digital Painting
Double Exposure
Dreamy Fantasy
Dystopian Art
Etching
Expressionism
Fauvism
Flat Design
Fractal Art
Futurism
Glitch Art
Gothic Art
Gouache
Greco-Roman Art
Impressionism
Ink Wash
Isometric Art
Japanese Ukiyo-e
Kinetic Typography
Lithography
Low Poly
Macabre Art
Magic Realism
Minimalism
Modernism
Monogram
Mosaic
Neon Graffiti
Neon Noir
Origami
Papercut
Parallax Art
Pastel Drawing
Photorealism
Pixel Art
Pointillism
Polyart
Pop Art
Psychedelic Art
Rennaissance/Baroque
Retro Wave
Romanticism
Sci-Fi Fantasy
Scratchboard
Steampunk
Stippling
Surrealism
Symbolism
Trompe-l'œil
Vaporwave
Vector Art
Voxel Art
Watercolor
Woodblock Printing
Zen Doodle
Hue
The image hue can be used to define a color overtone to the images. This can help you follow a color scheme or a common tone.
Below is an example of a cluster using Teal as the hue:
Troubleshooting
General information and advice
DALL-E 3 is a relatively new model from OpenAI and may still be a bit unstable. We think it will improve over time, especially as OpenAI adds more resources to run the service and perfects the platform but we have noticed a number of issues to be aware of.
I was charged for an image that was not generated
Machined makes use of timeouts and other mechanisms to ensure service stability. This can sometimes lead to us cancelling API requests before we've received a response from OpenAI (when they take too long to respond to us). This can sometimes lead to occasions when an image has not been generated but you have been charged for it on the API.
I used the same style but the image came out different
To ensure images are unique we provide a unique seed ot each image generation and as such, the output can be very non-deterministic. We strive to perfect our prompts to ensure as much consistency as possible but this may not always be possible. If you are unhappy with an image, you can always generate a new one.
Was this helpful?