EasyUse/Loadersgenerated

EasyLoader (Mochi)(mochiLoader)

Loads a Mochi checkpoint and VAE, encodes the provided positive and negative prompts, creates an empty latent with the requested dimensions, length, and batch s...

mochi Loader

model_override
clip_override
vae_override
pipe
model
vae
ckpt_name
COMBO
vae_name
mochi_vae.safetensors
positive
STRING
negative
STRING
resolution
width x height (custom)
empty_latent_width
INT
empty_latent_height
INT
length
INT
batch_size
1
Easy Use

Description

Loads a Mochi checkpoint and VAE, encodes the provided positive and negative prompts, creates an empty latent with the requested dimensions, length, and batch size, and returns an EasyUse pipeline along with the loaded model and VAE. It is intended for Mochi video generation workflows and is located in the EasyUse/Loaders category.

Inputs

NameTypeDefaultDescription
ckpt_nameCOMBOnoneMochi checkpoint file to load.
vae_nameCOMBOmochi_vae.safetensorsVAE file to load with the checkpoint.
positiveSTRING""Positive prompt used to guide generation.
negativeSTRING""Negative prompt used for CFG; leave empty for unconditional guidance.
resolutionCOMBOwidth x height (custom)Selection of preset output resolutions. Leave on width x height (custom) to use the manual empty_latent_width and empty_latent_height values.
empty_latent_widthINTnoneWidth of the generated latent / output frames.
empty_latent_heightINTnoneHeight of the generated latent / output frames.
lengthINTnoneNumber of frames to generate for the Mochi video.
batch_sizeINT1Number of video samples to generate in parallel. Min 1, max 4096.
model_overrideMODELoptionalSupply a model to use instead of loading one from ckpt_name.
clip_overrideCLIPoptionalSupply a CLIP text encoder to use instead of the checkpoint’s CLIP.
vae_overrideVAEoptionalSupply a VAE to use instead of vae_name.

Outputs

TypeDescription
PIPE_LINEEasyUse pipeline containing the loaded model, VAE, conditioning, empty latent, and generation settings for downstream EasyUse nodes.
MODELLoaded Mochi diffusion model.
VAELoaded Mochi VAE, useful for decoding latents to video.

Usage Notes

  • This node is designed for Mochi video generation: set length to the desired number of frames, then use the MODEL output with standard sampling nodes for video.
  • positive and negative are text strings. They are encoded internally, so this node does not expose a separate CLIP output.
  • When resolution is left on width x height (custom), the manual empty_latent_width and empty_latent_height fields control the spatial dimensions.
  • The PIPE_LINE output is the most convenient connection point for other EasyUse nodes; the MODEL and VAE outputs are available when you need to wire standard ComfyUI nodes directly.
  • Memory usage scales with batch_size, length, and the chosen width and height, so increase these values conservatively.
  • Use the optional override inputs when you already have a loaded model, CLIP, or VAE in your workflow and want to avoid reloading files.

Comments

Sign in with GitHub to join the discussion.

Loading comments…