Full prompt for Astra xhigh 3D forest build shared
The author shares the full prompt used on GPT-6 Astra xhigh, which asks the model to spend 5 to 24 hours building a dense real-time 3D forest in the browser and to keep checking vegetation, lighting and performance.
yall asked for the prompt, here it is: (ran it on GPT 6 Astra xhigh) -- First search for the "/unlazy" skill, read it fully, and use it throughout the entire task. Treat this as a major visual build lasting at least 5 hours. Prefer 8–24 hours of meaningful implementation and iteration if the environment allows it. Do not stop after producing one decent-looking scene. Create an extremely detailed real-time 3D forest in the browser. Nothing else. No landing page, cards, story, marketing UI, dashboard, or unnecessary text. The forest itself is the entire experience. Push density and visual quality extremely hard. Build massive amounts of detailed grass, varied trees, branches, leaves, bushes, ferns, moss, flowers, weeds, roots, fallen wood, stones, soil, leaf litter, small plants, terrain variation, and tiny environmental details. The forest should feel almost excessively dense. Empty ground should be rare. Focus obsessively on: realistic tree silhouettes, dense grass, natural asset variation, rich colors, lighting, contact shadows, wind, foliage motion, atmospheric depth, fog, sun shafts, particles, terrain, materials, scale, composition, and detail at both close and far distances. Avoid obvious repeating assets, billboard-looking foliage, tiled terrain, floating plants, plastic materials, identical trees, sparse procedural placement, excessive bloom, or a generic Three.js demo appearance. Include smooth free-camera exploration. Repeatedly capture and inspect the scene from many positions. Identify weak vegetation, empty areas, bad colors, repetition, poor materials, unrealistic lighting, ugly tree shapes, performance bottlenecks, and anything that breaks immersion. Fix these issues and continue increasing quality throughout the run. The goal is not feature count. The goal is to make one browser forest look completely ridiculous in terms of detail and atmosphere. Build the full experience, publish it PUBLICLY on ChatGPT Sites, verify the deployment, and return the public URL.
This post quotes another post: x.com/lexnlin/status/2096263046918197609