Help with LTX-Video 2.3 SDR-to-HDR on 16GB (or Dual 16GB)
Hi everyone,
I'm trying to run a high-fidelity SDR-to-HDR workflow using LTX-Video 2.3, but I’m hitting constant OOMs and, if I try a lighter text encoder (like gemma 3 4B) I get "invalid tokenizer" errors on a 4070 Ti Super (16GB) and even a dual 16GB (4070 Ti S + 5060 Ti) configuration.
Does anyone have a stable configuration or a "lite" text-encoder setup that successfully runs on a single 16GB GPU, or offloads to the second GPU? Any tips on fixing the tokenizer pathing for sharded 4B models in the LTX custom nodes would be a lifesaver.
Thanks!
Setup
4070 Ti Super 16GB or dual 16GB (4070 Ti S + 5060 Ti 16GB)
64gb ram
You can use the free API for prompt encoding:
https://github.com/Lightricks/ComfyUI-LTXVideo/blob/master/gemma_api_conditioning.py
Thank you.
Just for clarification: is it normal that it doesn't work on a 16gb card? Is there a way to make it work locally with this setup?
Thanks again for the help.
16gb Vram is low but with optimizations (quantized models, tiled vae etc...) you can get things to pass, how much frames are you trying to get through the model? you might want to try the looping sampler with small enough chunks of frames.
I have seen people using 16gb in banodoco discord, just In general when you have such low amount of VRAM you have to spend time optimizing things.
Thank you. I'm pretty new to this, so I may not understand all you said, but I'll look into it. How much ram is recommended? Thanks again
@ilm1429hotmail You can try this workflow and try to make it work on your computer : https://civitai.com/models/2566341/ltx-23-22b-ic-lora-outpaint-workflow-gguf-low-vram
It's a workflow that uses another ic lora not the HDR one, you'll need to make 3 changes mainly :
- Switch outpaintI lora to the HDR LoRA
- Skip the node that pads/resizes the input video for outpainting
- Add the LTXV Logc3 decode node at the end for EXR output
I haven't tried that workflow by the way, so if you encounter specific issues you might need to reach out to its author.
Thank very much for your time and help. I appreciate it. I shall try this right now :)
@ilm1429hotmail -- a great place to get help for this issue is our Discord server, there's regular discussions of running low-hardware configurations there:
https://discord.gg/ltxplatform
Sorry for the late reply. Thank you for the discord link!