The LTX Video Generation Node for InvokeAI enables users to generate dynamic videos using the LTX-Video pipeline from Hugging Face Diffusers. This node supports both text-to-video and image-to-video generation
This node requires diffusers version to be installed from source, to Install it using the following command:
pip install git+https://github.com/huggingface/diffusers
Please note that this may cause other issues in invoke (so far I have not noticed any)
| Field | Description |
|---|---|
| task_type | Select the generation task type: text-to-video or image-to-video. Default is text-to-video. |
| prompt | Text prompt for the video. |
| negative_prompt | Negative prompt to avoid unwanted artifacts, such as "worst quality, inconsistent motion, blurry, jittery, distorted." |
| input_image | Input image for the image-to-video task (ignored for text-to-video). Default is None. |
| width | Width of the generated video in pixels (selectable from predefined values, e.g., 128 to 1280). Default is 640. |
| height | Height of the generated video in pixels (selectable from predefined values, e.g., 128 to 1280). Default is 640. |
| num_frames | Number of frames in the video (selectable from predefined values, e.g., 9 to 257). Default is 105. |
| fps | Frames per second for the generated video. Default is 24. |
| num_inference_steps | Number of inference steps for video generation. Default is 30. |
| guidance_scale | Guidance scale for classifier-free diffusion. Higher values = stronger prompt adherence, lower values = better image quality. Default is 3.0. |
| seed | Seed for reproducibility. Use -1 for random behavior. Default is 42. |
| max_length | Maximum length of the input prompt in tokens (higher values may result in longer encoding times). Default is 256. |
| output_path | Path to save the generated video. Default is a generated_videos directory in the current file's path. |
| save_last_frame | Option to save the last frame of the video as an uncompressed PNG file. Default is False. |
| apply_compression | Enable compression artifacts to simulate video-like input. Default is False. |
| compression_intensity | Compression intensity level (higher = more compression artifacts, 0 = none). Default is 20. |