Pinned Tweet
OpenClawとVeo3.1で作ったCrimsonClawの予告編トレイラー🦞 稲妻→タイトル→都市降臨→飛翔まで4カット32秒。音楽もAI生成。 #VEO3 #CrimsonClaw #OpenClaw
1
1,620
A fictional guitar legend, brought to life with MiniMax H3 in ComfyUI. 🎸
Five connected 6-second segments, guided by a reference image and music, create a 30-second cinematic performance.
Warm light, quiet intensity, and a little AI magic.
#ComfyUI #MiniMaxH3 #aivideo
2
168
Ported my raytracer’s mushroom forest to one Shadertoy shader 🍄 Hundreds of mushrooms and trees come from repeating cells; the floating spores are a grid walked along each ray. Compiles fast, no textures. shadertoy.com/view/fXyXzw
1
92
Found a ComfyUI workflow that transfers a person’s movements from a video to a CG model. (The one on the right is 3D CG.)
1
1
176
I wrote a shader for each sushi topping in my own RenderMan-style shading language, compiled them with my own shader compiler, and rendered this scene with them. #Shadertoy #Renderman #Threejs
shadertoy.com/view/s3KXz1
You might not see what’s so impressive about it, but this real-time CG was completed in just a few minutes with a short prompt. I ported my own shader compiler for Three.js.
61
Angkor Wat: City of the Gods – 15-Minute Explainer Video
narration script + production notes (CG, animation, cuts, telops, music, direction)
Conceptualized with AI assistance (Copilot-style workflow for script, structure, and visual planning). Target runtime: ~15 minutes.
61
I created an intro CG of the Taj Mahal in about 15 minutes. CG modeling, rendering, script, editing, music, telop/subtitle creation, etc.—everything was done with Copilot.
45
A video like this can be made in 10-30 minutes. With just four photos, you can interview Einstein, Napoleon, and Shakespeare. I didn’t specify the language, so it came out in Chinese—but of course, I can also make them speak Japanese or English.
1
62
I tried having it do Japanese embroidery (Kyoto embroidery / Kyō-nui). Fable 5.1. Claude’s LLM has no drawing ability, but you can make it draw using SVG.
68
Astra's stunning visual power. The photorealistic illustration has been vectorized(svg) into approximately 690,000 paths, with no embedded images and fully editable text.
128





