Most creative teams try to automate repeatable creative tasks by retyping the same AI prompt every session, then watching that work vanish when the thread closes. The better way is to save the method once, then run it across a batch of assets you've already approved.

That's what a Skill does. It saves any AI edit as a named, reusable workflow with its own slash command. Think of it as a recipe: you bring the images, and the Skill brings the method. Once saved, it's there whenever you need it, and when you publish it to the workspace, everyone on your team can run it too.

Here's what's ahead: what a Skill is (in more detail) and what it isn't, how to build one step by step, four recipes you can copy today, and how a handful of saved edits becomes a shared library your team reuses.

A Skill is a named, reusable workflow saved from any AI edit you make in Canvas. You trigger it right from Canvas by typing "/" in the agentic chat. Select several assets on the canvas first, and the same Skill can run across the whole batch.

The sections below cover the three types of Skills you'll meet, how they handle brand context, and which jobs should use something other than Skills.

Built-in, private, and workspace Skills

Skills come in three tiers serving different scopes of work. Knowing which tier you're working in keeps your library clean and your team aligned.

• Built-in Skills ship with Air out of the box and need zero setup.

• Private Skills are ones a single user creates for their own repeat work, like a headshot crop only one photographer runs.

• Workspace Skills get published for the whole team to trigger.

Most teams have a few methods like that living in one person's head or an old chat thread. When that person is out, or leaves, the method goes with them, and whoever picks up the work starts from a blank prompt.

Publishing a Skill turns one person's method into a shared capability. Anyone with edit access in Air can create and run Skills, and publishing to the workspace makes a method official for everyone.

How brand context travels with every run

Air's Context is the brand memory layer behind Canvas. Each context source, such as a brand kit, stores your colors, fonts, logos, reference images, and usage instructions, and you can generate one from a website URL or add the details by hand.

Because brand context is already available to every Canvas edit, a Skill that uses a logo or reference image only has to name the Context source and the specific reference once, in its instructions. From then on, a Skill's output lands on brand without a separate setup step each run, which also means you don't have to hunt down your assets every time.

For a Marketing Lead adapting approved creative across channels, regions, and product lines, this removes a whole round of friction.

What Skills don't do

"Automate at scale" here means one specific thing: running a saved creative recipe across a batch of assets. It doesn't mean routing approvals or syncing systems.

For approval and status work, you can use Kanban views, custom fields, and approval status to handle that workflow inside Air. Zapier, Make, and the API can handle routing between tools. Outside agents working against your library are a separate capability, covered by Air's MCP server rather than by Skills.

Air's own guide is upfront that some edits aren't reliable yet. Spinning a product 360 degrees from a single photo forces the AI to invent the sides it can't see, so results improve a lot with four or more angle shots. Text wrapped around a 3D surface the model hasn't seen tends to come out inconsistent. And a very subtle edit, like a slight warmth shift, can get judged "good enough" with almost no change, so spell out the degree of change you want.

Basically, use Skills for volume work, and edit the handful of assets that carry the campaign by hand if you need to.

One more clarification: bulk resizing runs as a prompt to the Canvas agent, not as a saved or default Skill. Select your assets, tell the agent the ratio or dimensions you need (for example, "Resize all to 1:1 ratio"), and Canvas resizes the whole selection in one pass. This way, you can treat the resize prompt as your on-ramp to batch editing.

The path from a one-off Canvas edit to a named recipe your team runs on batches takes five steps. Each step adds reliability, from naming the Skill so people can find it to saving outputs where they won't get lost.

Step 1: Name the Skill and set its slash command

The name and "/" command are how teammates find and trigger the Skill, so both should describe the outcome, not the tool. A command like /brandmark-lowerright tells everyone exactly what happens. A command like /edit2 tells them nothing and ages badly.

For teams building several Skills, you should use a naming pattern to keep the library scannable as it grows. Try action plus placement or channel:

• /brandmark-lowerright

• /bg-swap-studio-white

• /localize-headline-ES

Consistent names mean anyone can open the "/" menu and read the library like a menu of outcomes.

Step 2: Choose the source asset type

Source asset type sets the kind of input a Skill accepts: images, video, both, or none. It controls what goes in, not the type of asset the Skill produces. It can't tell a product shot from a lifestyle photo, so that distinction belongs in the recipe's input check in Step 3.

Picture a background-swap recipe written for cutout product shots. If you run it on a group lifestyle photo, you might not get the results you expect since the recipe assumes a single isolated subject on a clean plate, not five people in a scene.

Source asset type only sets whether the Skill takes images, video, both, or none, so it can't tell a cutout product shot from a group photo. The input check in your recipe (Step 3) is what stops that mismatch before it starts, which saves a surprising amount of cleanup later.

Step 3: Write the recipe: goal, input check, steps, output

The Skills recipe has four parts. Writing each as its own line keeps the Skill readable and predictable. Here's a worked example for a watermark Skill you can adapt:

1. Goal: Add the brand wordmark to the lower-right corner. Do not change product shape, label copy, or image proportions.

2. Input check: Confirm the asset is a finished product or lifestyle shot. If it's a flat graphic or already watermarked, skip it.

3. Steps: Use the edit operation to apply the logo labeled [logo label] from the [Context source name] Context source at the set size and opacity. Preserve the original photograph; do not regenerate the image.

4. Output: Return the same dimensions as the source, one file per input, in the original format.

The goal has to state what must not change. Constraints are what keep batch output consistent across dozens of files. The input check then acts as a gate that tells the Skill what to verify before it edits and what to do when an asset doesn't qualify.

The steps should name the edit operation so the real photograph is preserved rather than regenerated, and the output line should specify format, dimensions, and count.

Step 4: Run the Skill on a batch in Canvas

Running a Skill on a batch of assets is the moment the setup time pays off. To batch edit approved assets with AI, multi-select the files in Canvas and trigger the Skill once. The same method applies across every selected asset in a single pass, as long as the recipe says to. Air's guide to writing Skills recommends adding a line like "If multiple images are provided, process each one independently and deliver one output per image." Without it, the AI may only process the first image.

Keep in mind this one habit to keep batch editing safe: check two or three outputs against the goal's constraints before you save the full set. For example:

• Did the product shape hold?

• Is the label copy untouched?

• Did the mark land where it should?

A thirty-second spot check catches a bad run before it multiplies across a hundred files.

Step 5: Save the outputs to the right place

Skill results stay on the canvas until someone saves them. They aren't added to a board automatically, and teammates who can see your canvas can view them but can't edit or save them, so make saving the last deliberate step of every batch.

Batch Skill outputs are new assets, not new versions. When you save them, they land as fresh items on the board you choose rather than stacking under the original, so don't go hunting for variants in a version stack that never received them.

For batch output, a save convention keeps everything findable later:

• A dedicated board per campaign or channel

• Custom fields for campaign, channel, and approval status

• A consistent naming scheme so filters actually work

With those in place, a marketer can filter to "Q4 holiday, Instagram, approved" and self-serve without pinging a designer.

The four recipes below show how to automate creative production with reusable AI workflows across the highest-frequency launch work: brand marks, background swaps, localized copy, and mockups.

The first three are recipes you write yourself using documented Canvas agent workflows. The fourth already ships as a built-in Skill, so it needs no recipe writing at all. Copy the structure, adjust the constraints to your brand, and you've got a starter library.

1. Brand mark or watermark across a full batch

Bake the default placement, size, and opacity into the instructions so the mark lands the same way every time, and add a note that placement can be overridden inline for individual assets. That way a product shot with a busy lower-right corner can take the mark elsewhere without breaking the Skill.

Here's the constraint to write into the goal: never cover the product, the logo lockup, or legal copy. This keeps a batch mark from ruining the shot it's supposed to protect.

An example of a use case for this recipe is an ecommerce drop where every product and lifestyle shot needs the same mark before it ships to retail partners.

Multi-select the whole shoot, run the Skill, spot-check three files, and save the set to the campaign board.

2. Background swap across a product set

Write this recipe around the edit operation so the actual product photograph is preserved while only the background changes. This way the real hero object stays intact.

Here are two constraints to give the recipe:

• Keep shadows and reflections consistent with the new background

• Keep product scale and crop identical across every asset in the set

You could use this recipe for one shoot stretched across a season: take a single product shoot and adapt it to seasonal, promotional, and evergreen backgrounds without booking a reshoot.

3. Localized copy variants that keep the layout

Build this recipe around Canvas text editing, where copy updates in place on the image without the source file. Instruct the Skill to preserve type hierarchy, alignment, and safe margins as the words change.

The constraint to lock in: layout and brand type styles hold across every language variant, even when string length changes.

For example, German tends to run long, and the Skill needs to respect the safe margins rather than let a translated headline collide with the edge.

The use case here is a campaign key visual shipped across regions. Translate the headline into each market's language and keep the design intact, with no fresh design round per locale. The regional teams get on-brand variants, and the studio never touches the source file.

4. Out of Home Mock as a built-in Skill

This one already ships with Air, so there's no recipe to write yourself. Run /ooh-mock straight from the "/" menu, and Out of Home Mock returns three photorealistic placements by default: two billboards and a bus side.

You can use this Skill for pitching. Imagine: you're presenting a campaign to executives or partners and want billboard and transit context, not a flat JPEG on a slide. Pull an approved key visual, run the Skill, and hand over realistic placements in minutes instead of waiting on a mockup artist. It turns "imagine this on a billboard" into something people can actually see.

Teams don't usually jump straight to scaled creative. They climb a ladder, and it helps to know which rung you're standing on:

1. Scattered folders across Drive, Dropbox, and desktops

2. Organized assets in one visual workspace

3. An approved source of truth everyone trusts

4. Reusable edits saved as Skills

5. Scaled channel variants running on batches of approved work

Here's what to do in the next 30 days. Pick the edit your team repeats most, the one someone does forty times a launch. Write it as a Skill with explicit constraints, spelling out what must not change. Then publish it to the workspace and track how often it runs.

That's the one-to-many payoff you get with Air: a DAM with AI that goes beyond a traditional DAM, building on your best work instead of just storing it.

Open Air, save your first Skill, and run it across a batch of approved assets today.

What is a Skill in Air?

A Skill is a saved, named AI edit you create in Canvas and reuse whenever you need it. It comes with its own slash command and stays private to you until it's published to the workspace for your whole team. You trigger it by typing "/" in the Canvas agent.

Do I need to know how to write prompts to create a Skill?

No deep prompting skill is required, but a clear recipe helps. If you can describe your goal, name what must not change, and specify the output you want, you can write a Skill that holds up across a batch. The four-part structure of goal, input check, steps, and output does most of the heavy lifting for you.

Do Skills work on video assets?

Yes. When you create a Skill, you set its source asset type to images, video, both, or none, so a Skill can be built to take video clips as its input. The recipes in this article are written for image batches. For video-first edits such as trimming and image-to-video, Canvas's wider AI editing toolkit covers them directly.

Is this the same as templated creative automation?

No. Templated automation drops your content into fixed layouts. Skills run a saved editing method on your own approved assets, with brand context already applied. The output reflects real creative work rather than a filled-in template. You're multiplying assets you've already signed off on.

Who can publish a Skill to the whole workspace?

Anyone with edit access can create and run Skills. New Skills start out private to the person who made them, and users with the required management access publish them to the workspace from the Skills management center, turning one person's method into a shared team capability.

Which plans include Skills and credits?

Skills work on any plan for anyone with edit access, and AI generation draws from your plan's monthly credit balance. Credit amounts change over time, so check your current workspace plan for the live number rather than relying on a fixed figure.