Instructions to use RuneXX/LTX-2.3-Workflows with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use RuneXX/LTX-2.3-Workflows with LTX-2:

# Install the LTX-2 pipelines
git clone https://github.com/Lightricks/LTX-2.git
cd LTX-2
uv sync --extra natten

# Download weights from this repo
# Substitute filenames from this repo's "Files and versions" if they differ
hf download RuneXX/LTX-2.3-Workflows \
    diffusion_models/<distilled-transformer>.safetensors \
    text_encoders/gemma4-12b-with-proj-ltx-2.5-bf16.safetensors \
    vae/<video-vae>.safetensors \
    vae/<audio-vae>.safetensors \
    latent_upscale_models/<spatial-upsampler>.safetensors \
    latent_upscale_models/<temporal-upsampler>.safetensors \
    --local-dir models/LTX-2.3-Workflows
# DFR requires the detailing IC-LoRA (separate repo; strength is fixed at 0.5)
hf download Lightricks/LTX-2.5-22b-IC-LoRA-Pixel-Spatial-Upscaler --local-dir models/LTX-2.5-22b-IC-LoRA-Pixel-Spatial-Upscaler

# Distilled LTX-2.5 pipeline (fast)
uv run python -m ltx_pipelines.distilled \
    --transformer-path models/LTX-2.3-Workflows/diffusion_models/<distilled-transformer>.safetensors \
    --text-encoder-path models/LTX-2.3-Workflows/text_encoders/gemma4-12b-with-proj-ltx-2.5-bf16.safetensors \
    --video-vae-path models/LTX-2.3-Workflows/vae/<video-vae>.safetensors \
    --audio-vae-path models/LTX-2.3-Workflows/vae/<audio-vae>.safetensors \
    --spatial-upsampler-path models/LTX-2.3-Workflows/latent_upscale_models/<spatial-upsampler>.safetensors \
    --num-frames 121 \
    --prompt "A beautiful sunset over the ocean" \
    --output-path output.mp4
# For image-to-video, add: --image path/to/image.jpg 0 0.8

# DFR pipeline (higher detail fidelity; optional temporal 2x/4x)
uv run python -m ltx_pipelines.dfr_pipeline \
    --transformer-path models/LTX-2.3-Workflows/diffusion_models/<distilled-transformer>.safetensors \
    --text-encoder-path models/LTX-2.3-Workflows/text_encoders/gemma4-12b-with-proj-ltx-2.5-bf16.safetensors \
    --video-vae-path models/LTX-2.3-Workflows/vae/<video-vae>.safetensors \
    --audio-vae-path models/LTX-2.3-Workflows/vae/<audio-vae>.safetensors \
    --spatial-upsampler-path models/LTX-2.3-Workflows/latent_upscale_models/<spatial-upsampler>.safetensors \
    --temporal-upsampler-path models/LTX-2.3-Workflows/latent_upscale_models/<temporal-upsampler>.safetensors \
    --detailing-lora models/LTX-2.5-22b-IC-LoRA-Pixel-Spatial-Upscaler/ltx-2.5-22b-ic-lora-pixel-spatial-upscaler-x2-1.0.safetensors \
    --spatial-upscalings 1 \
    --temporal-upscalings 1 \
    --height 1088 \
    --width 1920 \
    --num-frames 121 \
    --prompt "A beautiful sunset over the ocean" \
    --output-path output.mp4
# For 4K: --spatial-upscalings 2 --width 3840 --height 2176
# For image-to-video, add: --image path/to/image.jpg 0 0.8

Notebooks
Google Colab
Kaggle

last frame keep changing from the original inputted last frame

#73

by Tyfani890 - opened Mar 31

Discussion

Tyfani890

Mar 31

plz tell me how to fix this, my last frame always not accurate from the inputed frame

gpundt

Mar 31

yeah same I had a someone holding a bottle of water for the last frame input and the words on the bottle were significantly different than the last frame that I input and I had a custom input audio and it would not lipsync not matter what I prompted

Tyfani890

Mar 31

i think the problem is on upscale part i got the video i wanted by skipping the upscale part

RuneXX

Owner Apr 1

What workflow?

gpundt

Apr 1

What workflow?

I was using the FL2V Custom audio workflow and having trouble with the last frame having different fine details from the input last frame, but I never tested further to see if it was actually the spatial upscaler was causing it like Tyfani890 said.

RuneXX

Owner Apr 1

•

edited Apr 1

2-pass (2 samplers) will generate more potential differences than 1-pass. So potentially the upscale part can create slight differences from input frame to output.
Its also a strength setting per input frame, you can try set it to 1.0 if its not already.

And I'll check that particular workflow. LTX has 2 ways of using reference frames... Either by a guider or an in-place* ... the guider, is as the name suggest a guide. And the model has freedom to not be a "slave" to the frame inputs (this usually gives a more natural looking video). The in-place is more strict frame injection, that can lead to more accurate re-representation of the input frames, but the video can end up looking a bit jerky or not as smooth, all depending .. .

Also the latent upscaler model had a last frame bug, so make sure you use the latent upscaler model v1.1

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment