Higgsfield demos GPT-6 Astra operating Photoshop, Procreate, and Figma via MCP

• Higgsfield turned one of its images into a fully editable Figma poster through its Astra and MCP setup, according to Higgsfield's Figma poster demo.
• Direct drawing demos put GPT-6 Astra at a Wacom Cintiq in Photoshop and on an iPad in Procreate, as Higgsfield's Photoshop run and Higgsfield's Procreate run show.
• Adobe work can split between UI control and code: icreatelife's script workflow says a model-written script kept token use low while the host application did the heavy processing.
• A reference-video recreation can retain its parts as separate editable compositions, per Higgsfield's After Effects run.

OpenAI’s launch post includes a house modeled in Blender and made walkable in Unreal. Its computer-use documentation describes the two paths visible in these clips: structured mouse and keyboard actions, or model-written code executed in a supplied environment.

The Figma example starts with a Higgsfield image, then moves into a native poster file through “GPT-6 Astra + Higgsfield MCP + Figma.” Its layers remain editable after the run.

A separate desktop demo has the agent operate a Wacom Cintiq 22, enter Photoshop through a Higgsfield plugin, and paint a comic-book cityscape. The Procreate clip in the TL;DR follows the same physical-control pattern on an iPad.

ChatGPT’s embedded Adobe connections cover simpler functions in icreatelife's explanation, while a more complex Photoshop task used a generated script instead. That division is the practical detail behind the slick cursor demos.

icreatelife said the script itself used very few tokens, leaving Photoshop, Illustrator, After Effects, or Premiere to execute the work without consuming generation credits.

Higgsfield’s 20-minute After Effects recording breaks the reference recreation into three steps:

1. Analyze the reference video.
2. Recreate motion elements to its timing.
3. Place every element in its own editable composition with a transparent background.

The same account’s cartoon workflow ran a kids’ cartoon through Higgsfield, edited it in Premiere, then uploaded the result to YouTube.

One of the more concrete 3D workflows started with roughly 400 photos from 2019. bilawalsidhu ran them through RealityScan, which its official documentation describes as a photo-to-3D reconstruction tool, then imported the photogrammetry mesh into Blender for Astra to clean up, texture from the scan, and extend with procedural shaders.

The reconstruction inherited some errors from the source mesh. A follow-up added a critic that compared the Blender output against multi-view imagery in two loops of about 45 minutes each, with a longer overnight pass planned afterward.