AE GPT documentation.

Introduction

AE GPT is an AI agent and project-aware chat panel for Adobe After Effects. It runs inside the Adobe Extensions panel, reads useful project context, helps debug expressions, writes ExtendScript, uses references in supported modes, accepts voice prompts, and can execute scriptable After Effects tasks through Agent Mode.

The normal setup path is simple: install AE GPT, activate the license, connect one AI provider, choose Chat Mode or Agent Mode, then add references or voice input when the task needs more context.

Installation

1. Download AE GPT from aescripts and install it with the aescripts + aeplugins Manager App, or install the ZXP package with your preferred ZXP installer.
2. Restart After Effects or Premiere Pro after installation.
3. Open the extension from Window > Extensions > AE GPT.
4. Sign in or activate your license if prompted by the aescripts licensing flow.

Open the panel

In After Effects or Premiere Pro, choose Window > Extensions > AE GPT. You can dock, resize, or float the panel like any other Adobe extension.

If the menu item does not appear, restart the Adobe app, reinstall the extension, and confirm that the ZXP was installed into the correct CEP extensions directory.

API connection

API key mode is the most direct way to connect cloud AI providers. Open AE GPT settings, choose Gemini, OpenAI, or Claude, select API mode where available, paste the provider key, save, and pick a model. Local mode uses Ollama instead of a cloud API key.

CLI connection

Supported CLI modes let AE GPT use provider command-line tools that are already authenticated on your machine. This can be useful when you prefer account-based login over direct API keys.

Install Node.js first so npm is available. On macOS, users who prefer package-manager installs can use Homebrew where the provider supports it.

Official setup references: Codex CLI guide Claude Code install guide Gemini CLI guide.

1. Install and authenticate the provider CLI outside Adobe first, such as Codex CLI, Claude Code CLI, or Gemini CLI.
2. Restart After Effects or Premiere Pro so the extension can detect the updated environment.
3. Open AE GPT settings, select the provider, choose CLI mode, and click the login or check button.
4. Select a model from the refreshed list and send a small test prompt.

node -v
npm -v

OpenAI / Codex CLI

Use this when the OpenAI provider is set to CLI mode. AE GPT checks the codex command and uses your local Codex login. For the latest official instructions, open the OpenAI Codex CLI install guide.

npm install -g @openai/codex
codex login
codex login status

Claude Code CLI

Use this when the Claude provider is set to CLI mode. Install Claude Code, open it once in Terminal, finish the sign-in flow, then restart After Effects. Check the Claude Code installation guide and Claude CLI reference when provider commands change.

npm install -g @anthropic-ai/claude-code
claude --version
claude

Gemini CLI

Use this when the Gemini provider is set to CLI mode. Run gemini once in Terminal and complete the Google sign-in flow before using it inside AE GPT. Google maintains the current install steps in the Gemini CLI get-started guide.

npm install -g @google/gemini-cli
gemini --version
gemini

macOS Homebrew alternatives

brew install --cask codex
brew install --cask claude-code
brew install gemini-cli

CLI attachment support can be more limited than API mode. If image or PDF references are required, switch to a supported API mode.

Local Ollama

Local mode detects Ollama on your computer and uses installed local models. This is useful for private prompting and offline-style workflows where you do not want to send prompts to a cloud provider.

1. Install Ollama.
2. Download any local model you want to use.
3. Open AE GPT settings, choose Local, and AE GPT will list the downloaded models in the Local model section.
4. Select a model and send a small test prompt.

ollama pull llama3.1
ollama pull qwen2.5-coder
ollama list

Voice input

Voice input lets you dictate longer creative instructions. AE GPT first uses provider transcription when a supported API key is available. If provider transcription is not available, CLI and local modes can use local Whisper on your computer.

Install local voice input

py -m pip install -U openai-whisper
winget install -e --id Gyan.FFmpeg

python3 -m pip install -U openai-whisper
brew install ffmpeg

Pre-download a Whisper model

Whisper downloads the selected model the first time it is used. If you want to see the download progress in Terminal first, run one of these model commands.

python -c "import whisper; whisper.load_model('tiny')"
python -c "import whisper; whisper.load_model('medium')"
python -c "import whisper; whisper.load_model('large')"

Chat Mode

Use Chat Mode when you want explanations, planning, troubleshooting, expression help, or project-aware answers without running an Agent Mode task. Premiere Pro can be used for chat-style assistance, but AE GPT does not directly edit video or audio for you.

• Ask about the active After Effects composition or selected layers.
• Ask Premiere Pro workflow questions as a bonus chat workflow, not as direct timeline execution.
• Debug expressions and ask for safer rewritten versions.
• Keep different ideas separated with chat history.

Agent Mode

Agent Mode is for After Effects tasks that AE GPT can turn into a plan and execute as scriptable steps. It is best for building compositions, creating shape layers, organizing layers, applying effects, writing expressions, and automating repeatable work.

The output quality depends heavily on the selected model. Stronger models usually write better ExtendScript, make clearer plans, and create cleaner After Effects results. Smaller, cheaper, or local models can still be useful, but they may produce weaker scripts or need more correction.

1. Open an After Effects project and choose the right comp or selection.
2. Switch the composer mode to Agent.
3. Write a specific prompt with comp size, style, duration, and constraints.
4. Review the visible step cards while AE GPT works.
5. Undo in After Effects if the result is not what you wanted, then refine the prompt.

Images and PDFs

Supported API modes can use images and PDFs as reference context. Attach moodboards, UI screenshots, client notes, storyboard frames, or technical documents when the visual direction matters.

• Supported attachments are limited by provider and mode.
• Local Ollama and some CLI modes may not see attachments.
• Keep reference files focused and avoid uploading private credentials or sensitive client data.

Choose a model

Pick a model based on the job. Strong reasoning models are useful for complex ExtendScript and Agent Mode plans. Faster models are useful for quick iteration. Local models are useful for privacy-first prompts.

Your first prompt

Start with a specific request and name the scope. For example, say whether AE GPT should work on the active composition, selected layers, or a new composition.

In the active composition, create a new 1920x1080 scene.
Add a dark solid layer as the floor, create one editable shape-layer button,
center it in the frame, and name every layer clearly.

AE GPT works best when large ideas are built step by step. Instead of asking for a full sci-fi scene in one prompt, first ask it to create the holographic character, adjust the position and colors yourself, then ask for the shields, then ask for the animation. This gives the model less to guess and gives you more control over the final After Effects comp.

Troubleshooting

Most setup problems come from the selected provider, a missing command-line path, or Adobe not seeing a newly installed tool until it restarts.

node -v
npm -v
where codex
where claude
where gemini
npm config get prefix

[Environment]::SetEnvironmentVariable("Path", $env:Path + ";$env:APPDATA\npm", "User")

echo 'export PATH="$PATH:/opt/homebrew/bin:/usr/local/bin:$HOME/.npm-global/bin:$HOME/.codex/bin:$HOME/.claude/local:$HOME/.gemini/bin"' >> ~/.zshrc
source ~/.zshrc

ollama list
ollama pull llama3.1

py -m pip install -U openai-whisper
winget install -e --id Gyan.FFmpeg

Public discussion

Use the discussion page for public questions, issue reports, workflow examples, and answers from other AE GPT users. Search first, then post with your Adobe app version, provider mode, selected model, and the exact visible error if one appears.