Breaking

Higgsfield Shows GPT-6 Astra Generating a Scene and a TouchDesigner Overlay

September 6, 2026 at 03:29 EDT

Foundation Models
Media Generation
AI Agents

Higgsfield AI has shown OpenAI's GPT-6 Astra assembling a detailed motion graphic in two stages: the model generated the scene inside Higgsfield, then built an animated tracking overlay in TouchDesigner. The company shared a 15-second clip of the finished look, with a generated scene followed by a glowing object-tracking graphic locked to motion in the frame.

Sep 6, 2026 · Higgsfield AI

A Scene, Then a Tracking Overlay

GPT-6 Astra generated a detailed visual in Higgsfield, then built an animated tracking overlay in TouchDesigner — two tools, one agent-driven pipeline.

How the clip was built

1 · GENERATE

Higgsfield scene

Astra builds the base visual inside Higgsfield.

→

2 · OVERLAY

TouchDesigner nodes

Astra constructs an animated tracking graphic.

→

3 · RESULT

Motion graphic

Scene plus a glowing lock-to-subject overlay.

3D Jutsu message cost

Approximate credits per agent message. Column height equals credits. Auto is free.

Auto
Free

GPT-5.6 Sol

~11

Fable 5.1

~17

GPT-6 Astra

20%

fewer tokens vs models Higgsfield tested, per CEO Alex Mashrabov

15s

length of the shared Higgsfield clip

1–60s

3D Jutsu scene time; default is 15 seconds

What landed

The overlay looked production-smooth

Developers said the tracking graphic felt unusually locked, and that the two-tool combo read more like a small crew than a chatbot.

What still holds

It is still a split pipeline

The clip is a company demonstration. Astra is the costliest 3D Jutsu LLM, and long agent jobs can still burn millions of tokens.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.