Local Ai
Is there per-workflow analog of '--fp16-unet' cli option?
The search results did not return the specific Reddit thread content. Based on what is available from the search results and general knowledge of the topic, here is a factual summary for the knowle...
The search results did not return the specific Reddit thread content. Based on what is available from the search results and general knowledge of the topic, here is a factual summary for the knowledge base entry:
In ComfyUI, the global --fp16-unet CLI flag forces the UNet to run in FP16 precision at launch but applies server-wide with no native per-workflow override. Users seeking per-workflow UNet precision control can work around this by using the "Load Diffusion Model" node, which exposes a weight_dtype selector (e.g., fp16, fp8_e4m3fn, default) directly within individual workflows. This allows different workflows to load the UNet at different precisions without requiring a server restart or changes to the global CLI arguments.
Related
- b8756
- b8749
- b8748
- I can finally run LTX Desktop after the last update.
- What is the 'Unload Models and Execution Cache' from the ComfyUI menu doing that all the other model and cache-clearing nodes I've tried don't do?
Source: local-ai