Testing MiniMax-H3 T2I on RTX 5090: 12 Mins for a 15s Clip
VirtualWishX · reddit · 2026-08-07
The author tested the Text-to-Video (T2I) capabilities of the MiniMax-H3 model (FL2VA pruned int8) on an RTX 5090 (32GB) system.
- Performance: Generating a 15-second, 0.6-megapixel video at 20 sampling steps took about 12 minutes and 27 seconds.
- Objective: The author intentionally ignored the recommended complex prompt formats (like specifying Subjects and Pictures) to test the model's tolerance and creativity with "freestyle" natural language.
- Prompt Details: The post includes a long, camera-language-heavy prompt describing a cinematic action scene with Arnold Schwarzenegger and Jerry Seinfeld in a white void.
More from Infra
- MiniCPM-V 4.5 MoE reportedly hits 13+ tokens/s on iPhone with just 3GB memory — heyshrutimishra · 2026-10-16
- Solo-built Vulkan inference engine LaForge hits ComfyUI-level SD1.5 speed: 512px in 0.8s, 402MB container — Ichidan256 · 2026-10-16
- Trading 2 RTX 5090s for 4 R9700s: does doubled VRAM pay off? — poofph · 2026-10-16
- Running PolyStrata + Cline coding agent on a $0-fancy laptop at ~90 tok/s — rpwoerk · 2026-10-16
- TensorFold 1.0.12 runs 120B vision model on 1-2 DGX Sparks, adds Mac image input — HankYeomans · 2026-10-16
- vLLM brings frontier model local deployment to NVIDIA RTX Spark PCs — vllm_project · 2026-10-16