Reddit Asks for Real-World LongCat-Video Inference Times on RTX 4090 to H100
Clean_Extreme_3970 · reddit · 2026-09-22
A Reddit user asks for real-world inference numbers for the open-source LongCat-Video model (including Avatar 1.5), rather than official benchmarks.
They want to know how long it actually takes to generate 10s or 60s of finished video on RTX 4090, RTX 5090, L40S, A100, or H100, plus resolutions used and whether long videos are generated continuously or stitched from clips. The focus is purely GPU inference time and hardware requirements, not hosting costs.
More from Multimodal
- Princeton's VideoGen-Agent Adds 19 Points via Agentic RL Tool Use for Video — princetonu · 2026-09-22
- Intel Ships Day-0 OpenVINO Support for Qwen-Image-2.1 — Alibaba_Qwen · 2026-09-22
- Pexo Hands-On: Chat Your Way From Rough 15s Demo to 30s SpaceX Concept Film — SimplyAnnisa · 2026-09-22
- H3 Motion Context node causes flickering at clip starts, reports Reddit user — superdeathkillers · 2026-09-22
- CLIP-powered image search: 25,000 images in under 50ms, running locally in your browser — jason_mayes · 2026-09-22
- Pixel-art tour of London generated with MiniMax H3 shows consistent stylized video — Sufficient_Cause_43 · 2026-09-22