VFX Studio Complains: ComfyUI's Memory Management Causes Nightly OOM Crashes
vfxthrowawaydotcom · reddit · 2026-08-27
A VFX studio opens a professional discussion on Reddit about ComfyUI's unreliable memory management in production. The studio runs RTX 6000 Pro GPUs on Windows, with workflows chaining 2-6 model steps (Z-Image, Qwen-Klein, MiniMax, LTX), requiring constant model load/unload.
Typical failure: rerunning the exact same job 40 times, the first 4 succeed, then iteration 5 randomly OOMs. Artists waste significant time rebooting ComfyUI; MiniMax nodes have been error-prone since release.
They've tried launch flags, clearing VRAM/RAM between steps, and profiling model sizes — none fully working. The studio is now considering building its own DCC tool (with Claude's help) but would prefer to stay on independently developed ComfyUI, and asks how larger studios achieve long-term stability.
More from Multimodal
- Open-source LoRA turns MiniMax H3 into seamless 360-degree orbit videos from a single photo — camenduru · 2026-10-15
- Neon Noir Elegance prompt template goes viral with striking pink-blue two-tone outputs — LudovicCreator · 2026-10-15
- Krea 2 Turbo + px16 LoRA generates game-ready pixel art in 5s, where GPT Image 2.5 fails — Angaisb_ · 2026-10-15
- How Goldfish Made Its Kung Fu Scenes: Real Nanquan Choreography + Seedance 2.5 — JunieLauX · 2026-10-15
- Custom px16 LoRA makes Krea 2 Turbo churn out pixel icons at 5s each — Angaisb_ · 2026-10-15
- Prompt share: 'Crimson Footsteps' makes one red element pop in monochrome AI photos — azed_ai · 2026-10-15