MiniMax shares creator's 2D-to-3D video made with H3

"Connecting 2D to 3D" This is the trail of my efforts, wondering how far I could go if I layered ideas onto a generated video. Top left is the original generation, top right is the finished piece, and below is the Blender screen. ■What I did 1. Create a KV in Midjourney 2. Generate a 10-second run with MiniMax H3 3. Keep the same motion but change only the material to make 7 versions (paint / chrome / powder / ink / flowers / water / thread) 4. For the first half's 2D, use Python to switch materials per pose 5. Build a 3D space in Blender that looks the same as the 2D, and connect them 6. In 3D, the camera moves into the trails of the materials 7. Pulling back out, the fragments gather into an image of a giant runner 8. Add light streaks and floor reflections inside Blender as well ■What I learned ・Combining generation with tools raises quality ・Connecting 2D to 3D gives a somewhat richer expression Maybe it's enough to get cool materials straight out of AI generation, but I hope this kind of approach serves as a reference too.

View original

「2Dを3Dに繋ぐ」 生成した動画にアイデアを重ねたらどこまでやれるんだろう、と頑張ってみた軌跡です。 左上が元の生成、右上が完成、下がBlenderの画面。 ■やったこと 1. MidjourneyでKVを作る 2. MiniMax H3で10秒の走りを生成 3. 同じ動きのまま、素材だけ変えて7本に（絵の具／クローム／粉／墨／花／水／糸） 4. 前半の2Dは、Pythonでポーズごとに素材を切り替える 5. 2Dと同じ見え方の3D空間をBlenderに作り、つなぐ 6. 3Dでは、素材の軌跡の中にカメラが入っていく 7. 抜けて引くと、欠片が集まって巨大な走者の絵になる 8. 光の筋や床の映り込みも、Blenderの中で足す ■わかったこと ・生成とツールを組み合わせるとクオリティUP ・2Dを3Dに繋げば多少リッチな表現に AI生成で元からかっこいい素材を出せばいいのかもしれませんが、こういったアプローチもありそうだなと参考まで。