- name: (STRING), ELLA model to load.
- ella: (ELLA), ELLA model.
- name: (STRING), T5 text encoder model to load.
- max_length: (INT), max_length of encoder, set
0
for flexible token length. - dtype: (BOOLEAN)
- t5_encoder: (T5_TEXT_ENCODER)
- model: (MODEL), loaded by
Load Checkpoint
and other model loaders. - ella: (ELLA), ELLA model loaded by
Load ELLA Model
node. - positive: (ELLA_EMBEDS), currently needing both.
- negative: (ELLA_EMBEDS), currently needing both.
- sigmas: (SIGMAS), Use
BasicScheduler
and same config withKSampler
to get sigmas. Fallback into legacy method(deprecated) if not provided and timesteps is not set.
- model: (MODEL), sd model with ELLA model injected.
- positive: (CONDITIONING), for
KSamplers
. - negative: (CONDITIONING), for
KSamplers
.
- ella: (ELLA), ELLA model loaded by
Load ELLA Model
node. Need to useSet ELLA Timesteps
first. - embeds: (ELLA_EMBEDS)
- conds: (CONDITIONING), for
KSamplers
.
If clip is provided to this node, it will automatically concat ella condition and clip condition.
- ella: (ELLA), ELLA model loaded by
Load ELLA Model
node. Need to useSet ELLA Timesteps
first. - text_encoder: (T5_TEXT_ENCODER)
- text: (STRING), prompt to encode.
- clip: (CLIP), clip model to encode text_clip.
- text_clip: (STRING), prompt to encode with clip.
- conds: (CONDITIONING), for
KSamplers
.
- text: (STRING), prompt to encode.
- text_encoder: (T5_TEXT_ENCODER)
- embeds: (ELLA_EMBEDS), for chain combination ELLA_EMBEDS
- ella_embeds: (ELLA_EMBEDS)
- embeds: (ELLA_EMBEDS)
- embeds_add: (ELLA_EMBEDS)
- embeds: (ELLA_EMBEDS)
Deprecated, 'Combine CLIP & ELLA Embeds' instead
use for convert clip CONDITIONING
to ELLA_EMBEDS
- cond: (CONDITIONING)
- embeds: (ELLA_EMBEDS)
Deprecated, 'Combine CLIP & ELLA Embeds' instead
- cond: (CONDITIONING)
- embeds: (ELLA_EMBEDS)
- embeds: (ELLA_EMBEDS)
- model: (MODEL), loaded by
Load Checkpoint
and other model loaders. - ella: (ELLA), ELLA model loaded by
Load ELLA Model
node. - scheduler: (SCHEDULER_NAMES), need to be consistent with
Ksamper
. - steps: ("INT"), need to be consistent with
Ksamper
. - denoise: ("FLOAT"), need to be consistent with
Ksamper
.
- ella: (ELLA), ELLA model with timesteps injected.