SDXL Base and Refiner: a workflow does not always need both

File and download boundary: ComfyUI World provides source references and metadata only. Model and LoRA weights, node packages, and workflow files stay with their original providers; this site does not store, mirror, or proxy them. Downloads and access conditions are controlled by each provider.

Purpose and scope

This reference explains the roles in official SDXL examples, not a benchmark of every derivative checkpoint or a claim that every LoRA is interchangeable.

Source-supported context

Official examples support both base-only and base/refiner paths. Component mismatches can cause dimension failures, so the template determines whether refinement belongs in the graph. Source 1 Source 2 Source 3

Before you start

1. For a base-only baseline, confirm checkpoint, text encoding, latent and VAE roles.

2. In a refiner workflow, identify the handoff stage and its conditioning rather than adding an arbitrary second checkpoint.

3. Consult a derivative model's own card/version details; the SDXL label does not describe all internal choices.

Steps and checks

The sequence below is an editorial procedure based on the cited context, not a diagnosis already confirmed for your environment.

Step 1. Select the official base-only or base/refiner example that matches the task and retain its original artifact and source.

Step 2. Identify each checkpoint's role and encoder pairing, ensuring separate loaders do not accidentally select the same download.

Step 3. Test a small baseline before adding refinement, LoRA or ControlNet, checking family compatibility for every addition.

Step 4. Record template-to-weight relationships. For shape/CLIP errors, return to that chain rather than substituting an SD1.5 encoder.

Completion check

Each model has a clear purpose and optional branches have an understood effect. Adjust quality parameters only after a small task passes.

Limitations and cautions

This record does not invent unchecked SDXL filenames, sizes, hashes or VRAM figures. Follow the resources linked by the official example and the individual model card.

Original sources

• SDXL examples — original record checked 2026-09-21.
• ComfyUI model troubleshooting — original record checked 2026-09-21.
• LoRA workflow — original record checked 2026-09-21.

Localization prepared: 2026-09-22. The source-check dates above were inherited from the existing article; this translation does not claim they were all rechecked today. No GPU run, minimum-memory measurement or runtime guarantee is supplied.

Related troubleshooting and guides

• LoRA has no effect or reports size mismatch: check the base model and both output branches
• mat1 and mat2 shapes cannot be multiplied: inspect the model component chain