Minimax H3 Latent Upscaler slows to 1000+ seconds per step when queueing jobs
WhiteKnight225 · reddit · 2026-08-27
A Reddit user testing the Minimax H3 latent upscaler reports that the first generation runs normally, with each of the upscaler's 3 steps taking roughly 60–180 seconds. From the second queued generation onward, however, the upscaler initialization and each step balloon to 1000+ seconds. The poster is asking whether others hit the same issue and whether a fix or setting exists — likely a resource-release problem under multi-job queuing, but no conclusion yet.
Related event: Minimax H3 Latent Upscaler Slows Sharply in Queued Generation(2 posts)→
More from Multimodal
- Google releases Omni 1.1 Flash with 4K output and frame interpolation — legit_api · 2026-08-27
- Minimax H3 generates greetings from 20 women in 20 dialects — SIR_NVAX_A_LOT · 2026-08-27
- How to generate synthetic character data for LoRA with limited originals? — BeautifulStation4 · 2026-08-27
- Krea Launches H3 Max: 15 Seconds of Video Generated in 5s, Claimed 50x Faster — isidentical · 2026-08-27
- AI-Generated Short 'The Gifted' Makes Future Vision XPRIZE Top 50 of 2,500 Entries — LudovicCreator · 2026-08-27
- VFX Studio Complains: ComfyUI's Memory Management Causes Nightly OOM Crashes — vfxthrowawaydotcom · 2026-08-27