Kentskooking-ComfyUI-BridgeLab/docs/node-reference.md

Node Reference

SD1.5 To SDXL Embedding Converter

Purpose: Convert one SD 1.5 textual inversion embedding into an SDXL .safetensors embedding.

Inputs:

• sd15_embedding: source embedding from the ComfyUI embeddings model list
• sdxl_checkpoint: target SDXL checkpoint from the ComfyUI checkpoints model list
• output_name: optional output filename stem
• overwrite: allow replacing an existing output file
• run_ci_validation: run phrase-based validation after fitting the alignment
• validation_phrase_count: number of sampled phrases used during validation
• validation_batch_size: validation batch size

Returns:

• embedding_name
• embedding_path
• fit_token_cosine
• validation_hidden_cosine
• validation_pooled_cosine
• report

Behavior:

• Reads the source embedding through load_embedding_bundle().
• Rejects anything that is missing CLIP-L, has a CLIP-L width other than 768, or already contains CLIP-G.
• Builds or reuses a checkpoint-specific alignment cache.
• Saves the converted embedding into the active ComfyUI embeddings directory.

SD1.5 Folder To SDXL Embedding Batch Converter

Purpose: Convert every supported embedding in a folder into SDXL outputs.

Inputs:

• input_folder_path: source directory to scan
• sdxl_checkpoint: target SDXL checkpoint
• name_suffix: suffix appended to each output embedding stem
• recursive: include nested folders
• overwrite: allow replacing existing outputs
• run_ci_validation: run one shared alignment validation step before conversion
• validation_phrase_count: number of sampled phrases used during validation
• validation_batch_size: validation batch size

Returns:

• output_directory
• converted_count
• skipped_count
• failed_count
• fit_token_cosine
• validation_hidden_cosine
• validation_pooled_cosine
• report

Behavior:

• Writes outputs under models/embeddings/converted_sd15-SDXL.
• Reuses one alignment fit and one optional validation run for the whole batch.
• Preserves relative folder structure after sanitizing path components.
• Skips or fails items independently and records each outcome in the report.

SD1.5 To SDXL Hypernetwork Converter

Purpose: Project a supported legacy SD 1.5 hypernetwork into SDXL-sized branches.

Inputs:

• sd15_hypernetwork: source hypernetwork from the ComfyUI hypernetworks model list
• output_name: optional output filename stem
• overwrite: allow replacing an existing output file

Returns:

• hypernetwork_name
• hypernetwork_path
• report

Behavior:

• Loads the source payload onto CPU with comfy.utils.load_torch_file().
• Converts only classic linear payloads accepted by convert_linear_sd15_hypernetwork_to_sdxl().
• Writes the converted payload into the active ComfyUI hypernetworks directory as a .pt file.

Common Failure Modes

• The selected checkpoint does not expose both SDXL CLIP-L and CLIP-G token embedding tables.
• The source embedding is already SDXL or is not a standard SD 1.5 textual inversion embedding.
• Validation is enabled but the bundled CLIP Interrogator phrase files are missing.
• The output file already exists and overwrite is disabled.
• The hypernetwork payload uses unsupported features such as non-linear activation, layer norm, or dropout.