This is the only LTX 2.3 video workflow so far I that I can run on my NVIDIA GeForce RTX 3050 4GB vram GPU and 8GB ram laptop.
It's not fast on 4GB and probably much faster on 6GB and 8GB machines.
I recommend having these settings in your run_nvidia_gpu.bat file:
--windows-standalone-build --lowvram --force-fp16 --use-split-cross-attention
So change the first line to something like this:
.\python_embeded\python.exe -s ComfyUI\main.py --windows-standalone-build --lowvram --force-fp16 --use-split-cross-attention
Here are the filenames and where to Place them. The files can be found on huggingface.
C:\AI\ComfyUI_windows_portable\ComfyUI\models\unet
ltx-2-3-22b-dev-Q4_K_M.gguf
C:\AI\ComfyUI_windows_portable\ComfyUI\models\clip
gemma_3_12B_it_fp4_mixed.safetensors
ltx-2.3_text_projection_bf16.safetensors
C:\AI\ComfyUI_windows_portable\ComfyUI\models\vae
LTX23_video_vae_bf16.safetensors
LTX23_audio_vae_bf16.safetensors
C:\AI\ComfyUI_windows_portable\ComfyUI\models\latent_upscale_models
ltx-2.3-spatial-upscaler-x2-1.0.safetensors
C:\AI\ComfyUI_windows_portable\ComfyUI\models\loras
ltx-2.3-22b-distilled-lora-dynamic_fro09_avg_rank_105_bf16.safetensors
If you change the frame rate, do it on both video and audio so it stays synced.
Description
Can use some optimization
FAQ
Comments (9)
How much system RAM you need?
With a 8GB RTX 3070 and 16GB of RAM (on Linux and with an Intel integrated taking care of the desktop, so most RAM and VRAM is available) I can only get it to make 768x512 for 4 second videos (and not every time...)
Hi. I can do 1024x768 with my 4GB RTX 3050 and 8GB of RAM. I can do 6 seconds at 20fps stable. So 5 seconds at 24fps I guess. I have done longer but it might crash. It should not be a problem for you.
Sorry I think I missunderstood. You can only get it to make 768x512 for 4 second videos with this workflow? That is strange. It takes forever for me but it doesn't crash.
@Benkebengt Thank you dear friend, you are talented
@Benkebengt yeah, sometimes 5s if I'm lucky but then I get an out of memory. Maybe it's because I have CUDA 13.2 and/or because of the latest changes in memory management in ComfyUI? I'm running it with --listen --force-fp16 --use-split-cross-attention --disable-dynamic-vram --disable-pinned-memory --lowvram
@TooLittleVRAM @TooLittleVRAM @TooLittleVRAM I've been getting more out of memory errors too lately so might be an ComfyUI update. Trying to figure out a way to make it use less memory.
works great for the most part, takes around any wear from 10-20 minutes render time depending on picture size. My only problem, could you do an update that includes sampler previews?
didn't work on 4G VRAM(16G RAM), but did work on 6G VRAM(32G RAM). thanks
Is this workflow still working?