Launch args to run LTX?

Custom Node Testing

• I have tried disabling custom nodes and the issue persists (see how to disable custom nodes if you need help)

So how does a Linux user actually run LTX 2.3 nowadays, any of the official Comfy workflows? 7900XTX, 64GB RAM. All workflows crash due to filled system RAM to 99% -> OOM, even with 768x512 res, 100 frames.

Support tells users to "stop disabling our optimizations" but even Klein crashes on first gen if i don't disable "smart-memory-management." At least i can generate images with that launch arg, but LTX is a no-go no matter what i try.

All this started at around v0.17-0.18, before that i could run LTX 2.3 just fine, even at 1000 frames.

Something happened.

[coderabbitai]

⚠️ Possible Duplicate Issue(s)

• Short LTX 2.3 Workflow that yesterday worked now provokes oom killer #13023

---

🧪 Issue enrichment is currently in open beta.

You can configure auto-planning by selecting labels in the issue_enrichment configuration.

To disable automatic issue enrichment, add the following to your .coderabbit.yaml:

issue_enrichment:
  auto_enrich:
    enabled: false

💬 Have feedback or questions? Drop into our discord!

[patientx]

This is about those optimizations vs AMD gpu's (specially older ones, before 9000's)

try these (on windows they enable me to generate ltx at higher resolutions - frame counts not sure about linux)

--disable-api-nodes --cache-none --disable-smart-memory --disable-pinned-memory --disable-dynamic-vram --reserve-vram 16

yes I reserve all vram, and yes the memory before these "optimizations" wan-ltx was at least possible now too slow or oom'ing

[jorismak]

my intel system has this too.

some options, try to disable anything called 'dynamic-vram' for as long as we still can (cause it can be awesome on some systems, it destroys comfyui on most others).

or switch to gguf types, because that loader bypasses the whole dynamic-vram part.