Higgsfield integrates GPT-6 Astra motion design into After Effects

• Higgsfield's ChatGPT workflow takes uploaded assets and an idea, then assigns the manual work to Astra, according to Higgsfield's launch post.
• Aspect-ratio versions are the first narrowly defined feature: Higgsfield's reframing demo says a composition can be converted into platform-native formats in one pass.
• A 3D brain became an editable After Effects setup through a prompt-built GPU plug-in in Higgsfield's 3D brain demo, while the butterfly project claims 600 editable Photoshop layers after six hours of work.
• A first-run DaVinci Resolve test by moritzkremb got files, a project, cuts, and overlays assembled, but judged the initial output basic.

Higgsfield's MCP guide divides the work cleanly: Astra codes, reasons, and orchestrates; Higgsfield supplies creative assets and its generation/deployment workflow. The plugin page puts installation in ChatGPT's Plugins Directory, and OpenAI's model documentation lists five Astra reasoning-effort settings, from low to max.

The public material presents @Higgsfield /use-after-effects as a path from a brief to agent work inside Adobe After Effects. It describes three inputs:

• Uploaded assets, an idea, and direction over the visual treatment.
• A product website, used as reference for a commercial with editable camera movement, lighting, typography, and animation.
• A bespoke GPU plug-in. Higgsfield says its neural-connections demo built one from a prompt, with each connection individually animatable.

Higgsfield's reframing demo starts from an existing After Effects composition and generates aspect-ratio variants for platform-specific delivery.

Effects 2.0 uses a separate /effects command: upload a photo in ChatGPT, then generate a short effects video in the chat.

The announcement calls the feature free, while a Higgsfield reply says the free trial includes one generation.

The first public test outside Higgsfield's own demos came from moritzkremb, who gave Astra access to DaVinci Resolve. It retrieved files, created a project, and made edits, but the result remained basic in that initial run.

moritzkremb found basic cuts worked better than visual finishing: their follow-up says overlays and animation were harder. A second test sent the same raw footage to Tella, DaVinci Resolve, and Hyperframes, and their verdict preferred the Hyperframes edit.

Higgsfield's butterfly test is a production recipe, not a one-shot render. The company says it spent five to six hours creating 120 frames across 600 editable Photoshop layers.

1. Generate a butterfly-in-flight image sequence with GPT-Image 2.5, as Higgsfield describes.
2. Feed handmade animation samples into the process to study paper and drawing-material behavior, according to Higgsfield's thread.
3. Build custom charcoal and pencil Photoshop brushes, which Higgsfield says it made for the project.
4. Draw the 120 frames layer by layer over five to six hours, as Higgsfield reports.

Higgsfield's examples repeatedly use reference material to construct editable 3D scenes rather than only producing a final clip.

• A flat artwork becomes a textured Blender model in the 2D-to-3D demonstration.
• A classic skeleton dance becomes a 3D Blender animation in the skeleton-dance clip.
• A few office photos become a navigable room after the system analyzes references and infers missing details, according to the office-reconstruction demo.
• A SketchUp demonstration claims a five-minute build of a furnished estate, pool, tennis court, and grounds from a $100 million villa brief in Higgsfield's villa demo.
• The Blender integration places generated objects into a scene in Higgsfield's plugin clip.
• A CLO3D brand pipeline moves from garment sketches to 3D clothes, campaign imagery, and a storefront in Higgsfield's clothing-brand demo.

The demos also extend the motion-design workflow into technical visualization.

• The Houdini crowd demo combines running characters, collisions, and ragdoll dynamics into a pileup.
• A side-by-side test compares Astra and DeepSeek 4.1 Flash on motion tracking and data labeling in Higgsfield's comparison.
• The bridge demo models the Brooklyn Bridge in AutoCAD, then claims to analyze structure, materials, and load-bearing capacity with tanks crossing from each direction.
• A traffic workflow uses traffic-camera data to model light-timing alternatives, then reconstructs the streets in Blender, according to Higgsfield's traffic simulation.
• A flight explainer visualizes airflow, lift, and thrust through animation in Higgsfield's physics lesson.
• A neural-connections video pairs a brain visualization with an Astra-versus-Fable comparison in Higgsfield's brain clip.

Higgsfield is publishing model comparisons alongside full creative workflows, without attaching a numerical result to the visual tests.

• The commercial prompt specifies a 30-second school-courtyard story, comic-book effects, fast camera moves, a product shot, and a slogan.
• A separate motion-design comparison puts Astra and Fable 5.1 side by side.
• The collage workflow generates artwork piece by piece, assembles it in Photoshop, then animates it in After Effects.
• A self-insert film demo lets the subject shape a story and publish it, as shown in Higgsfield's movie clip.

Higgsfield's published After Effects sequence is:

1. Open ChatGPT and install Higgsfield from Plugins.
2. Type @Higgsfield /use-after-effects.
3. Create with Astra in computer-use mode inside Adobe After Effects.

OpenAI says Astra is rolling out across ChatGPT Plus, Pro, Business, and Enterprise, plus its API, Azure, and AWS Bedrock in its launch announcement. Higgsfield's help center says MCP and other automated surfaces deduct standard credits, while the company's trial reply specifies one free generation for the trial.