Skip to content
Back to home
AI 3D

A photoreal forest, from Megascans and one photo

Michiyoshi Shirota wondered if GPT-6 Astra, given only Megascans assets and one reference photo, could assemble a photoreal forest in Blender. It could. He suspects it might not have needed to be Astra.

last month

While everyone was posting Astra-in-Blender modeling, Michiyoshi Shirota tried a blunter idea. Skip the sculpt. Hand the model Megascans assets only, plus one reference image, and ask for a forest. Photoreal, he thought, might fall out of the library rather than out of a generated mesh. The result, he says, is pretty good.

Then he undercuts his own demo: maybe this did not have to be Astra. That sentence is why the post belongs here. A lot of clips exist to crown a model. This one is a hypothesis about inputs. A scanned twig is already halfway to a photograph. The model’s job is taste, placement, and light.

One reference image is a very small brief for a forest. The clip is how much of the photograph was already hiding in the asset pack.

A photoreal forest, from Megascans and one photo

Made with

Michiyoshi Shirota

@onihachi

View the original post on X
GPT-6 AstraでのBlenderモデリングが、劇流行りしてますが、そもそもmegascansのアセットだけを使って森を作ってもらえば、すぐフォトリアルになるんじゃないかと思ってやってみました。結構良い結果になりました。参照で渡したのは添付の画像だけです。これならastraじゃなくてもできるのかな。 https://t.co/bfyhqaaWhy
Translated by AIWow editors

1.4K likes

likes
1,428
replies
20

Captured when this was curated, not live. —

Keep reading