EasyLoader (SVD)(svdLoader)

The EasyLoader (SVD) node (svdLoader) loads an SVD-based Stable Video Diffusion checkpoint, its VAE and CLIP, takes an input image, and prepares an EasyUse PIPE_LINE for image-to-video generation. It also exposes the raw loaded MODEL and VAE.

Inputs

• ckpt_name (COMBO, required): Select the SVD checkpoint to load.
• vae_name (COMBO, required): Select the VAE used for decoding latents.
• clip_name (COMBO, required): Select the CLIP model to load into the pipeline.
• init_image (IMAGE, required): The source image used as the first frame and conditioning input for video generation.
• resolution (COMBO, required, default "1024 x 576"): Quick preset for the video resolution. Use it as a convenient starting point for common SVD aspect ratios.
• empty_latent_width (INT, required): Width of the empty latent used as the generation starting point.
• empty_latent_height (INT, required): Height of the empty latent used as the generation starting point.
• video_frames (INT, required, default 14, min 1, max 4096): Number of frames to generate.
• motion_bucket_id (INT, required, default 127, min 1, max 1023): SVD motion conditioning value. Lower values favor less motion; higher values favor stronger motion.
• fps (INT, required, default 6, min 1, max 1024): Frames per second of the generated video.
• augmentation_level (FLOAT, required, default 0, min 0, max 10, step 0.01): Amount of noise to add to the input image before diffusion. 0 keeps the init image unchanged; higher values allow the output to deviate more strongly from the original frame.
• optional_positive (STRING, optional, default ""): Positive prompt text carried through the pipeline.
• optional_negative (STRING, optional, default ""): Negative prompt text carried through the pipeline.

Outputs

• PIPE_LINE: An EasyUse pipeline containing the loaded SVD model, VAE, CLIP, init image, video settings, and optional prompt strings.
• MODEL: The loaded SVD model.
• VAE: The selected VAE.

Usage Notes

• This node is intended for SVD checkpoints. Standard SD1.5/SDXL checkpoints may not produce the expected image-to-video behavior.
• motion_bucket_id is a key SVD conditioning signal: low values produce nearly static clips, while high values produce more pronounced movement.
• augmentation_level is useful for generating variations of the input image. At 0, the initial frame stays close to the source; increasing it gives the model more freedom to reinterpret the frame.
• optional_positive and optional_negative are supplementary prompt strings passed through the pipeline. SVD conditioning is primarily image- and motion-based, so downstream nodes decide how to consume these strings.
• Keep resolution and empty_latent_width/empty_latent_height consistent with the selected SVD model’s expected resolution to avoid distorted latents.