Automating Adobe After Effects with GPT-6 Astra and Higgsfield
Chase AIgo watch the original →
the gist
Use the open-source Higgsfield Motion Designer plugin to enable GPT-6 Astra to control Adobe After Effects, allowing non-designers to generate motion graphics via natural language prompts and storyboard-driven workflows.
Workflow for AI-Driven Motion Design
The integration relies on the Higgsfield Motion Designer plugin, which acts as a bridge between the LLM and Adobe After Effects. The process avoids manual keyframing by using the model to execute scripts and computer-use actions to manipulate the software directly. Because rendering can take 15 to 25 minutes for short clips, the workflow emphasizes front-loading the design process to minimize iterative prompt loops.
Implementation Steps
- Storyboard First: Generate a visual storyboard using the model's image generation capabilities before attempting any animation. This ensures alignment on the visual direction and pacing, preventing wasted render time on unwanted outputs.
- Reference Integration: Provide the model with reference videos or screenshots of desired motion styles. The model analyzes these to extract motion design elements, which it then attempts to replicate or adapt into the new project.
- Structured Prompting: Use detailed prompts that define the duration, specific scenes (beats), and technical requirements. A sample prompt structure includes:
I want a 12-second whiteboard animation in Adobe After Effects using the installed Higgsfield After Effects integration. [Reference file path]. The beats are: [Scene 1 description], [Scene 2 description]. - Iterative Editing: Treat the initial render as a draft. Use follow-up prompts to adjust transition speeds and timing, as the model defaults to 2.5 to 3-second scenes which often feel sluggish. Be specific about which elements to change to avoid unnecessary re-renders.
Context
Traditional motion design in Adobe After Effects requires significant technical expertise and manual labor. This approach lowers the barrier to entry by offloading the technical execution to an LLM, allowing users to generate complex graphics without prior experience in the software. The author notes that while the process is powerful, it remains a human-in-the-loop system where the user must guide the model through storyboarding and specific refinement prompts to achieve high-quality results.