Breaking

Higgsfield and GPT-6 Astra Visualize 500 Years of New York in 3D

September 5, 2026 at 14:27 EDT

Media Generation
Foundation Models
AI Agents

Higgsfield AI has used OpenAI’s GPT-6 Astra to produce a 3D visualization spanning 500 years of New York, showing Manhattan grow from fields and forests into a dense city and then into an imagined future. The creative-suite company framed the sequence as a time machine for the borough: a continuous 3D view of urban growth followed by a speculative skyline, built with Higgsfield tools and OpenAI’s newest frontier model.

September 5, 2026 · Higgsfield AI

500 years of New York, blocked in 3D

Higgsfield paired GPT-6 Astra with 3D Jutsu to show Manhattan growing from fields and forest into a city, then an imagined future.

500

Years of New York visualized

~17

Credits per GPT-6 Astra message

20%

Fewer tokens vs models Higgsfield tested

1–60s

3D Jutsu scene duration (15s default)

Agent message cost (credits)

Same unit, drawn to scale. Auto is free and omitted. Taller stack costs more per chat.

Claude Opus 5

~11

Claude Fable 5.1

~17

GPT-6 Astra

From prompt to city fly-through

1 · PROMPT

Describe the scene; pick GPT-6 Astra or another agent

→

2 · BLOCK

Edit objects, lights, and cameras without a full regen

→

3 · EXPORT

Render video from chat, or save MP4, still, or GLB

What landed

Creators treat city-scale blocking, camera paths, and long Unreal builds as newly practical in a browser agent.

What still breaks

First passes can miss a floor plan, later edits can break a file, and city reconstruction is not unsupervised.

The limit that still holds

Long 3D jobs are practical to attempt. Unsupervised city reconstruction is not proven.

A week-long Manhattan Unreal build filled streets and Central Park in simplified form. Higgsfield has not published a historical-accuracy check of the 500-year New York sequence.

Continue reading

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