Ask for a title sequence and get the tracks, keys and controls behind it.

Ape is the creative agent built into Grapple. It reads the project you have open, plans the work and builds it with the editor's own tools: tracks, clips, text, shapes, keyframes, cameras and effects. When it finishes you have a preview and an ordinary project to keep editing.

Give Ape a specific job.

Ape works best with the kind of note you would give an assistant editor: what to make, where it goes, when it happens, and what must not change. It looks at the project before it builds, so you can refer to the clip, the composition or the moment by name.

"Add a lower third for Maya Chen, sound recordist. Bring it in just after she starts speaking, hold it for four seconds, then fade it out. Keep the name and the role as separate text layers, and don't touch the edit."

The result is a pair of text layers on their own track, with keys you can see in the Graph editor. If the name is misspelled later, it's a text edit. If the entrance is half a second late, drag the keys.

For longer pieces, choose one of Ape's skills from the panel. Each is a working method for a kind of film: documentary cold opens and chapter cards built from archive stills and interviews, fast music-driven editorial with kinetic type, and layered scenes built from separately generated artwork, placed at different depths for the camera to move through.

Working with Ape

The construction work, done with the editor's tools.

Ape places and trims footage, builds tracks and compositions, and creates text, shapes and image layers. It sets blend modes and mattes, orders layers, adds markers and notes, and sets audio levels and fades. That makes it useful wherever a piece has many small parts that follow the same rules.

A set of lower thirds
Twelve names and roles from a list, each timed to where that person first speaks in the transcript, with the same style and entrance.
A flipbook or stop-motion run
Imported stills scheduled as an image sequence at the timing you give.
A vertical version
A 1080×1920 composition with the main cut nested inside, ready for you to reposition.
A layered illustration
Separate background, subject and foreground images placed at different depths, with a camera set up to move through them.

Everything Ape makes is available in the editor afterwards. Change one title without asking for the whole set again, or give Ape a composition you started and ask it to finish a defined part.

Movement you can open in the Graph editor.

Ape writes keyframes with easing for layers, text, shapes and cameras, so a title that rises into place or a push through separated layers arrives as ordinary keys. It can animate text by word or glyph, move a camera with depth of field for a rack focus, and attach a layer to a drawn path so it travels the route facing forward.

It also uses the less obvious animation tools: swapping images over time for replacement animation, rigs and poses for characters made of separate parts, and live drivers so one element trails another with a spring.

For procedural movement, Ape writes a property script whose amounts are exposed as sliders, such as the frequency and size of a handheld drift. You change the amount without touching the rule.

Treatments with controls you can adjust.

Beyond the Effects panel's built-ins, Ape can build colour fringing, a glitch, a directional smear, temporal echo trails, a flash frame, fragment shatter, a particle dissolve, a stroke reveal, particle emitters for dust or sparks, and more. Each arrives with named controls in Properties that you can key.

When none of those fit, Ape writes a custom effect in GLSL or Python. Say which controls you expect to change, such as strength or scale, so the effect is useful when the notes come in. Try it on a short representative range before applying it across a sequence.

Expressions and custom effects

Generate what is missing, then compose it.

Ape can run the image, video and audio models you have connected. Ask for a background plate and the separate foreground elements that will sit in front of it, and it generates them as named assets and places them in the project. Because the pieces are generated separately, they can move independently when the camera moves.

Ape can also transcribe speech when the Whisper Small model is installed, read a corrected transcript, and set up a new provider model for you from that provider's documentation.

Follow the plan and check the result.

While Ape works, the panel shows its plan and marks each step as it finishes. Type a note to steer the run in progress, or choose Stop. Ape renders previews and contact sheets of its own work as it goes, to check framing and timing before moving on.

When it finishes, the Changes view lists what it did, and the Tool-call view shows every step in detail. Choose Review in Stage to compare the frame before and after the run. In History the whole run is one entry that expands into its individual changes.

If the direction is wrong, choose Undo run to reverse the entire run in one step. Grapple checks that edits you made since can be kept, and tells you if they can't.

Decide what Ape is allowed to do.

Ape can run terminal commands, for example to measure rendered frames. Choose how those commands are approved:

Ask for approval
Every terminal command waits for you to allow it.
Approve for me
Sandboxed commands run automatically; anything else asks first.
Full access
Commands run with unrestricted filesystem and network access, without asking.

Turn on Grapple Ultra in the model menu to let Ape hand independent parts of a large task to private child agents. The Subagents panel shows what each one is doing, and you can stop them individually. While Ape is changing the project, Undo and Redo pause so the two of you don't edit at once.

Choose the model behind Ape.

Pick Ape's conversation model in Settings > Ape. Grapple has presets for Anthropic, OpenAI, Google, xAI, DeepSeek and OpenRouter, which use an API key from that provider. Ape can also run through Codex or Claude Code installed on your computer, using your own OpenAI or Anthropic sign-in. Where a model offers it, an Effort setting trades speed for more thorough reasoning.

The model receives what Ape reads while it works, including project details, transcripts and the frames it renders to check its work. Provider usage is billed by the provider under your account. You can also run Ape on a chat model hosted on your own computer.

Model setup guide

Try it on footage you already know.

Pick a short clip and one job, such as a lower third or a trimmed opening. The getting-started guide takes you from import to an exported file.