MiniMax H3 Tested: Generating the Will Smith Spaghetti Meme in 5s
topamine2 · reddit · 2026-08-03
Following the release of the MiniMax H3 video generation model, numerous real-world tests have emerged in the community. A user successfully recreated the iconic "Will Smith eating spaghetti" meme, demonstrating the model's strong capability in handling complex facial distortions and physical interactions.
According to the testing parameters, the video was generated at 864x480 resolution for a 5-second duration. It took about 66 seconds to render on a consumer-grade RTX 5090 GPU, showcasing highly efficient local inference.
Related event: MiniMax Video Model Goes Viral with Stunning User Tests(61 posts)→
More from Multimodal
- Fan animation made in Blender with Claude Code (Opus 5.5) — Icy-Appointment3583 · 2026-10-16
- Making Blender animations with Opus, Astra and Argon: how you prompt still matters — chongdashu · 2026-10-16
- Tessellation-free geometric queries on displaced surfaces cut caustics variance 10x — ssh4net · 2026-10-16
- Complex multi-link CAD mechanism finally animates, dev celebrates the breakthrough — zetalyrae · 2026-10-16
- OpenMOSS releases MOSS-Image-SFT: Apache 2.0 text-to-image model on Hugging Face — bdsqlsz · 2026-10-16
- Hyperflow open-sourced: 49 sampling steps down to 8, ~3x faster video generation — mhdfaran · 2026-10-16