EasyLoader (Full)(fullLoader)

Description

EasyLoader (Full) is an all-in-one loader node from ComfyUI-Easy-Use. It loads a checkpoint, applies an optional LoRA, creates an empty latent, encodes positive and negative prompts, and returns all of these as separate outputs plus a combined PIPE_LINE for convenient use with Easy-Use pipeline nodes.

Inputs

Model and VAE

• ckpt_name: COMBO, required. Select the checkpoint to load.
• config_name: COMBO, required, default Default. Select a custom model configuration to use with the loaded checkpoint.
• vae_name: COMBO, required. Select the VAE to load.
• clip_skip: INT, required, default -2, range -24 to 0. Determines how many CLIP layers are skipped. More negative values skip more layers.

LoRA

• lora_name: COMBO, required. Select the LoRA to apply to the loaded model.
• lora_model_strength: FLOAT, default 1, range -10 to 10, step 0.01. Strength applied to the model weights.
• lora_clip_strength: FLOAT, default 1, range -10 to 10, step 0.01. Strength applied to the CLIP encoder.

Resolution and Latent

• resolution: COMBO, required, default 512 x 512. Preset resolution dropdown for the empty latent. Selecting a preset sets empty_latent_width and empty_latent_height.
• empty_latent_width: INT, required. Final width of the generated empty latent in pixels.
• empty_latent_height: INT, required. Final height of the generated empty latent in pixels.
• batch_size: INT, required, default 1, range 1 to 4096. Number of identical latent images to generate in the batch.

Prompt Encoding

• positive: STRING, required, default empty. The positive prompt text.
• positive_token_normalization: COMBO, required. Options: none, mean, length, length+mean. Controls how token embeddings are normalized during prompt encoding.
• positive_weight_interpretation: COMBO, required. Options: comfy, A1111, comfy++, compel, fixed attention. Controls how emphasis and attention weights in the prompt are interpreted.
• negative: STRING, required, default empty. The negative prompt text.
• negative_token_normalization: COMBO, required. Options: none, mean, length, length+mean.
• negative_weight_interpretation: COMBO, required. Options: comfy, A1111, comfy++, compel, fixed attention.

Optional Overrides and Stacks

• model_override: MODEL, optional. If connected, this model is used instead of loading from ckpt_name.
• clip_override: CLIP, optional. If connected, this CLIP is used instead of loading from the checkpoint.
• vae_override: VAE, optional. If connected, this VAE is used instead of loading from vae_name.
• optional_lora_stack: LORA_STACK, optional. Applies additional LoRAs in stack form, in addition to the main lora_name and strengths.
• optional_controlnet_stack: CONTROL_NET_STACK, optional. Applies a stacked list of ControlNets to the generated conditioning.
• a1111_prompt_style: BOOLEAN, optional. Enable this to interpret the positive and negative prompts using A1111-style prompt conventions.

Outputs

1. PIPE_LINE — A combined pipeline object containing the loaded model, VAE, CLIP, positive/negative conditioning, and generated latent for use with other Easy-Use pipeline nodes.
2. MODEL — The loaded or overridden model, with LoRA applied.
3. VAE — The selected or overridden VAE.
4. CLIP — The loaded or overridden CLIP.
5. CONDITIONING — The encoded positive prompt conditioning.
6. CONDITIONING — The encoded negative prompt conditioning.
7. LATENT — An empty latent generated from empty_latent_width, empty_latent_height, and batch_size.

Usage Notes

Use the PIPE_LINE output when connecting to other Easy-Use pipeline nodes to avoid wiring every component manually. The individual MODEL, VAE, CLIP, CONDITIONING, and LATENT outputs are available when you need standard ComfyUI connections.

For prompts written with A1111-style emphasis syntax, select A1111 under positive_weight_interpretation and negative_weight_interpretation, and enable a1111_prompt_style if you want the prompt parser to follow A1111 conventions more closely.

The resolution preset is a convenience for populating empty_latent_width and empty_latent_height; the actual latent dimensions used are the explicit width and height values passed to the node.