AI Tools#
Text#
ChatGPT#
can be found at “www.chatgpt.com”
OpenAI’s conversational AI, sibling model to InstructGPT
trained using Reinforcement Learning from Human Feedback (RLHF), and fine-tuned using Proximal Policy Optimisation
- Limitations
sometimes writes plausible-sounding but incorrect answers
sensitive to tweaks to the input phrasing or same prompt multiple times
often verbose and overuses certain phrases
usually guess what the user intended, instead of clarifying with the user
sometimes responds to harmful instructions
Copilot#
can be found at “www.copilot.microsoft.com”
Microsoft’s conversational assistant that integrates with Microsoft 365 ecosystem
can be used for document creation, data analysis, project management, and communication such as summarising Outlook inbox
- Copilot Chat
generates answers based on web content
uses Open AI GPT model as underlying model
Gemini#
can be found at “www.gemini.google.com”
Google’s multimodal LLM to handle text, audio, images and more
- Limitations
can have inaccurate responses on complex or factual topics
reflect bias in training data
can fail to show a multiple perspectives
may not respond to appropriate prompts, and give inappropriate responses to others
can be vulnerable senseless prompts
Grok#
can be found at “www.grok.com”
X’s AI assistant to create documents, write code, generate images, and search real-time
gets data and view from X trends
NotebookLM#
can be found at “www.notebooklm.google.com”
Google’s AI for research and thinking, built using Gemini models
summarise PDFs, websites, YouTube videos, audio files, Google Docs or Slides
can make audio overview for different sources
Perplexity#
can be found at “www.perplexity.ai”
Perplexity’s conversational AI that includes real-time web search capabilities
responds based on current Internet content
Image#
AI Face Swap#
can be found at “www.aifaceswap.io”
replace a person’s face with another’s in an image or video
support JPG, PNG, and WebP
- Limitations
facial features must be clear
must only have one face in the photo, multiple face swap is still in beta
Adobe Firefly#
can be found at “www.firefly.adobe.com”
Adobe’s creative generative AI models embedded in Adobe’s apps and Adobe Stock
generate image, video and sound effects, translate audio and video, and collaborate with Firefly Boards
CleanUp#
can be found at “www.cleanup.pictures”
Clipdrop’s AI inpainting for object removal
can remove object, defect, people or text from a photo
Craiyon#
can be found at “www.craiyon.com”
Craiyon’s AI text to art generator, previously known as Dall-E mini
provides square, landscape, and portrait aspect ratios
have photo, illustration, vector and raw image presets
- Limitations
can have biases on training data
Dall-E#
can be found at “www.chatgpt.com”
OpenAI’s GPT-3 with more parameters trained to generate images from text
latest Dall-E 2 generates more realistic and accurate images with better resolution
Ideogram#
can be found at “www.ideogram.ai”
Ideogram’s AI tool to create images, and integrate text into images
has CSV-driven feature to batch generation
has presets of vertical, square and horizontal between 1:3 and 3:1
Midjourney#
can be found at “www.midjourney.com” or “discord.com/invite/midjourney”
Midjourney’s AI to create image from text and image to video
Perchance#
can be found at “www.perchance.org”
a platform for creating and sharing random generators
most generators are text to image
Video#
Hedra#
can be found at “www.hedra.com”
content creation platform for image, video and audio
can make videos of different styles with full-body animated, realistic and dynamic background, including live avatars with lip-sync
can also add voiceovers and AI-generated speech
HeyGen#
can be found at “app.heygen.com”
text or image to video generator, can add avatars and video translations
can also convert audio to video by using a talking photo avatar
InVideo#
can be found at “www.invideo.io”
text to video generator, can create by just providing a URL
can also add multi-lingual AI actors
video editor provides real-time multiplayer editing
Runway#
can be found at “app.runwayml.com”
can generate contents from complex physics-based simulations to realistic renders
Sora#
can be found at “sora.chatgpt.com”
OpenAI’s platform to create content from text, image or video
SonoVid#
can be found at “www.sonovid.ai”, also called Veo 3
Google’s AI to create videos from text and image using Veo 3 model
Audio#
Adobe Podcast#
Audio Clean#
Beatoven.ai#
ElevenLabs#
Play.ht#
Podcastie#
SUNO#
TTSMaker#
Presentation#
Chalkie#
Gamma#
Napkins#
Automation#
Make#
can be found at “www.make.com”
no-code platform to connect applications and automate workflows
- Scenario
a collection of modules
build a scenario, test, and schedule to run
can also add tools, flow control and text parser
unless named specifically, a scenario will be auto named based on app
best to organise scenarios in a folder
can run scenarios at intervals, immediately or on demand
can have a scenario chain, one triggering others, but must be within the same team
Sub-scenarios: parent scenario -> bridge scenarios -> child scenario
a scenario can define its input so that others can send data when triggering it
scenario inputs can be mapped in modules
- Module
any element added within a scenario
App: a group of modules
Bundle: data collected from Make, can download JSON input/output bundles
has action, search, trigger, iterator, and aggregator modules
search modules cannot modify data
for Google apps connection, free and regular Gmail account will need to be reauthorised every 7 days
modules should be connected to pass information, and can be cloned with existing config
ACID modules support rollback features
JSON is used to pass information between modules
- Aggregators
combine multiple bundles into single one
aggregators map, select items, and reduce, combine the selected ones
Tools Aggregator: table, text and numeric aggregators
Array Aggregator: take fields from multiple bundles, and aggregate into single array
output can be grouped to produce different bundles
data-driven aggregations can help understand business and make decisions
- Iterators
can only extract an array and split into separate bundles
cannot be used as a trigger
without iterator, Make only processes the first item in an array
always check for empty array
- Triggers
first action that launches the scenario
each scenario can only have one trigger
Polling Trigger: execute scenario at intervals, has a clock icon
Instant Trigger: execute scenario as soon as data arrives, has a lightning icon, can also be configured like a polling trigger
if not explicitly indicated, a trigger will always be considered polling
Basic Trigger: to create custom trigger, and define output bundle structure
- Router
branch workflows onto separate paths
useful to process data towards separate outputs
will not consume any operations
can set the order of routes
- Filter
allow or restrict specific data
has basic, text, numeric and datetime operators
allow AND and OR rules
- Scheduling
define time and frequency to run the scenarios
always run manually and test before setting to schedule
- Operations
Make is priced on operations usage
an operation is used every time a module performs an action, even if it does not produce a bundle, i.e. 3 operations for 3 actions
error handler modules, rollback, break, resume, commit, ignore, filters and routers do not use operations
save operations by running a single module only to check output bundle
disable routes to isolate and for less operations when testing, data cannot pass but the connection between modules remain
in a sense, an operation is used for each API call
if a scenario executes, it will always consume at least one operation
- Check Run
shown when a scheduled scenario was executed, but only the trigger module was activated and it does not find any new data
can be viewed in a scenario’s history
provide insight of operations count to manage scheduling
active by default on each scenario
- Organisations
contain scenarios, teams, users and data
a user can belong to multiple organisations, and own as many organisations, but only have one free organisation
each organisation can have a different subscription plan
location cannot be changed after the organisation is created
- Blueprint
automation template with configurations and functions
allow to share scenarios with others outside of the organisation
composed in JSON format
- Data Types
text, number, date, boolean
Collection: a group of different data types as single unit, also called object
Array: same data type, can sort, extract a range, search, iterate, aggregate, and other advanced data extractions
entering different data type as specified will get an error, but will attempt to convert type if compatible
format of date depends on profile settings
- Mapping
mapping data from one module into another
a single operation will generate at least one bundle
when mapping an array item, first item will be chosen if no index is given
need to create custom data structure when connecting to external services not provided by Make
- Functions
make data transformations on fields easier
has general, math, text, binary, date and time, and array functions
scenario and functions use date/time from Organisation
get(): extract a value from an object with dot notation on the raw name
map(): extract values from an array, and transform them to create a new array, always return an arrayuse
get(map())to use the data frommap()in the next moduleusing
get()andmap()can save more operations than using aggregators and iterators, as functions can be mapped directly in the module
merge(): easiest way to deal with arrays of same structurespread functions over multiple set variable modules, test often, and then only start to optimise
- Variables
allow organisations to store data and reuse
System and Custom variables
- Webhooks
to receive notifications as soon as an event occurs from another application
allow to send data to Make over HTTP, usually an instant trigger
can be used to connect different apps, even external, to Make
has a unique URL to receive a request
modules tagged with “Instant” have webhooks implemented
requests are stored in a webhook queue if the scenario is inactive, sequential processing is enabled, or the scenario failed
cannot use one webhook in multiple scenarios
query string, header and body are used to share information
can automatically determines data structure from request
- Errors
organisations usage limit errors
handling and validation errors with data passed between the modules
API related errors
module timeout error (usually 40 seconds)
inconsistency error when multiple scenarios modify the same database
data is saved in the logs when an error occurs
a scenario can be deactivated when too many errors occur
Incomplete Executions: if enabled, data and blueprint from a failed scenario run will be saved, and continue to run, e.g. bundle 3 will be processed even if bundle 2 causes error
can retry incomplete executions automatically, manually or in bulk
Make will not save data for incomplete executions if data is set to confidential
- Error Handlers
connected to specific modules, and will only run when the module has an error
error handling route (empty dots) can have an error handler, one or more modules, or a combination of both
error handlers give more control over the scenario behaviour, and do not consume any operations
must be placed at the end of the error handling route
Break: only stop for the error bundle, and save an incomplete execution, can configure more on the retry
Ignore: ignore the error bundle and continue running for next bundles
Resume: can set a substitute value for a module output
Commit: only for ACID, stop the scenario and confirm changes to the module, further bundles will not be processed
Rollback: only for ACID, stop the run and revert all changes to the module, further bundles will not be processed
rollback is the default behaviour when there are no error handlers or incomplete executions are enabled
- Data Stores
database within Make, store data in the form of record composed of label and data
every record in the data store has the same structure
allow transfer of data between scenarios
always be consistent with data, and modifying will impact integrity and cause to have duplicates within the data store
Make allows 10 MB of storage for every 10,000 operations per month
up to 1,000 data stores per organisation, and minimum 1 MB of data store size
to restore deleted data, need to manually find data from the scenario’s execution history, and copy back to the data store