Directus AI Operations Bundle

A small bundle of Flow Operations which enable interaction with the OpenAI and Stability API's.

Note: To use these you will need an API Key for the respective services.

Tested with Directus 10.6.0

Installation

The package is published to npm: npm install directus-extension-ai-sdk-bundle

Manual Installation

• Download or fork the repository
• Install the requirements
  npm install
• Build the extension
  npm run build
• Move the entire project to your extension folder (only the package.json and dist folder are strictly necessary)
  mv ./ extensions/directus-extension-ai-operation-bundle
• Restart your Directus instance

Bundle Content

Stable Diffusion Operation

This operation allow you generate images from text using the Stable Diffusion models with the Stability.ai API taking a prompt as input and saves the resulting image to the file library returning its ID as output.

Output
An object containing the ID of the resulting image file.

1{ "id": "[UUID]" }

Configuration Options
I am by no means an expert in what any of these model specific settings do so please check out the stability docs for that.

1. Prompt This is the main text input used for generating the image.
2. Style style to use, default is cinematic
3. Negative prompt, default is blurry, bad
4. Width Width of the image (note: height * width must be <= 1 Megapixel)
5. Height Height of the image (note: height * width must be <= 1 Megapixel)
6. CFG Scale How strictly the diffusion process adheres to the prompt text (higher values keep your image closer to your prompt)
7. Steps Number of diffusion steps to run
8. Engine The model to be used (defined choices can be found in src/lib/constants.js
9. API Key Override Used to set an API Key to be used specificly on this operation.
10. Storage location by default this is set to local but can be changed to s3 for example if your using s3.

DALL-E Operation

Similar to the Stable Diffusion operation this allows you to generate images from text but using the OpenAI DALL-E models instead with the OpenAI API taking a prompt as input, optionally saves it to the local file library and returning both the original URL and local image file ID.

Output
An object containing the ID(s) and URL(s) of the resulting image file(s). For a single result:

1{ "url": "https://some.url", "id": "[UUID]" }

For multiple results:

1{ "url": ["https://some.url","https://some.other.url"], "id": ["[UUID]","[UUID]"] }

Configuration Options

1. Prompt This is the main text input used for generating the image.
2. Save To File Library Disable importing the image into the local storage allowing you to use the DALL-E result URL for other systems.
3. Number of pictures DALL-E allows you to generate up to 10 images.
4. Image Size Choices are 256x256, 512x512 or 1024x1024
5. API Key Override Used to set an API Key to be used specificly on this operation.

GPT-3 Davinci Operation

This operation allows text completion using the OpenAI Davinci models with the OpenAI API taking a prompt as input and returns the generated text.

Output
An object containing the generated text:

1{ "response": "Generated TEXT" }

Configuration Options
I am by no means an expert in what any of these model specific settings do so please check out the OpenAI docs for that.

1. Prompt This is the main text input used for text completion.
2. Temperature Controls randomness: Lowering results in less random completions.
3. Maximum Length The maximum number of tokens to generate. Requests can use up to 2,048 or 4,000 tokens shared between prompt and completion.
4. Top P Controls diversity via nucleus sampling: 0.5 means hald of all likelihood-weighted options are considered.
5. Frequency Penalty How much to penalize new tokens based on their existing frequency in the text so far. Decreases the models likelihood to repeat the same line verbatim.
6. Presence Penalty How much to penalize new tokens based on whether they appear in the text so far. Increases the models likelihood to talk about new topics.
7. API Key Override Used to set an API Key to be used specificly on this operation.

Configuration Hook

This hook ensures the existence of API Key fields in the Directus Setting. These global settings will be used by the individual operations if not overridden in its local configuration.