Higgsfield demos GPT-6 Astra controlling Figma and Adobe creative tools

• Native creative-software control is the point of these demos: Higgsfield’s editable Figma poster demo turns one of its images into an editable Figma file.
• The model is shown operating physical and touch input surfaces, with Higgsfield’s Photoshop and Cintiq demo placing it on a Wacom Cintiq 22.
• Video work extends to reusable motion assets: Higgsfield’s 20-minute After Effects recording says each recreated element lands in its own transparent, editable composition.
• Browser access reaches publishing workflows, as Higgsfield’s YouTube Shorts demo claims Astra researched trends, launched a channel, and published ten videos.

Higgsfield’s TouchDesigner tracking-overlay demo separates the job in two: generate a scene in Higgsfield, then build the animated overlay in TouchDesigner. Outside Higgsfield, a MindStudio account describes a separate run that researched, scripted, voiced, and edited a YouTube video from one open-ended prompt.

Photoshop and Procreate demonstrations place computer use on different creative surfaces.

The Photoshop post says Astra controls a Wacom Cintiq 22, connects through the Higgsfield plugin, and draws a comic-book night cityscape. Higgsfield’s Procreate demo moves the same pattern to an iPad, where it claims the model completed a comic-style portrait inside Procreate.

The design examples preserve editability inside the target application.

In a Figma reconstruction, Higgsfield listed this reference-rebuild sequence:

1. Draw the outline.
2. Add light and volume.
3. Build the face.
4. Add the emblem.
5. Finish the texture.

Higgsfield’s Illustrator export demo adds a filesystem edge case: the agent built the native document, then routed through Finder when the app’s file picker froze before exporting.

The After Effects demo is the most detailed account of timeline work in the set.

Higgsfield says the run:

1. Analyzed a reference video.
2. Recreated motion elements against its timing.
3. Put each element in a separate composition with a transparent background.

For color work, Higgsfield’s DaVinci Resolve demo says Astra took Apple Log 2 footage and adjusted contrast, shadows, and color in four minutes. Its cartoon-to-YouTube demo adds Premiere editing and YouTube upload to the same computer-use chain.

Higgsfield’s 3D examples turn task descriptions into scene construction and rendering steps.

The floor-plan workflow is presented as:

1. Start from one floor plan.
2. Generate scene code.
3. Build the location model in Blender.
4. Render it with Cycles.

Most Astra-made scenes had been static before the Cartwheel MCP added animation, according to the Cartwheel MCP demo. Cartwheel’s research repository says it trained hundreds of human-motion models across scales; a later Higgsfield exploded-camera demo specifies a Blender camera built from 1,025 modeled objects.

The widest demonstrations cross from app control into accounts, platforms, and public outputs.

Higgsfield says that run went online to research trends, launched a new YouTube channel, and published ten Shorts with custom titles and tags. A LlamaParse and Loom demo says Astra logged into LlamaParse with its default tools, assembled slides from screenshots, then used full computer and Chrome access to try Loom and Screen Studio.

A reply about browser access describes a separate Midjourney experiment with access to the user’s browser, moodboards, and Midjourney account. OpenAI’s safety overview says that, with the right tools and access, Astra can find unknown security flaws and develop exploits without a person guiding each step.