.jpeg)
User Manual
1. The online workflow uses the LTX-2.3-22b-dev-fp8 model, and the local version can be adjusted according to your hardware configuration. It is confirmed that the Q4.gguf model is compatible with 12GB VRAM.
2. Pay attention to the notes next to the model when adjusting it and ensure correct node connections.
3. KJ's model offers the best cost-performance balance for low-end hardware, fast generation speed and high quality. The GGUF version has the lowest hardware requirements, while the LTX-2.3-22b-dev-fp8 model delivers the best output quality.
4. There are specific rules for resolution selection, see the notes next to the resolution settings for details. Landscape mode is highly recommended; portrait mode is supported but will produce subpar results.
5. Test conditions: Generating a 10-second image-to-video at 1280×736 resolution on a test PC with RTX 5060 Ti 16G + 32G RAM.
- ltx-2.3-22b-distilled_transformer_only_fp8_scaled: 335s
- ltx-2.3-22b-dev: 20min
- LTX-2.3-22b-dev-fp8: 500s
- LTX-2.3-22b-dev-Q4_K_M.gguf: 430s
- LTX-2.3-22b-dev-Q8_0.gguf: 470s
6. The auto-generated prompts have been tested repeatedly and can also be used with prompt templates to generate video prompts in other AI agents.
7. Toggle the corresponding switch to enable the I2V text-to-image function.
Resolution Selection Guidelines
（1）The output video dimensions of LTX must be multiples of 32. For a strict 16:9 aspect ratio, refer to the following resolutions: 512×288; 1024×576.
（2）Other approximate 16:9 resolutions: 1920×1088.
（3）Input resolutions of 1280×720/736 will both output a video of 1280×704.
（4）1280×768 is a fully supported resolution for normal output.
（5）Higher resolutions yield better visual effects but require more powerful hardware configurations.
This workflow is available for free online testing—download it for local use only if you are satisfied with the results.
Test URL: https://www.runninghub.ai/post/2030943530058784769/?inviteCode=rh-v1443
Bonus Points: 1000 points for new registrations, 100 points for daily login. A single 10-second 1280×736 generation consumes only 80 points.