Model Basics

ComfyUI Flux vs SDXL vs SD1.5: VRAM Model Choice

Choose between Flux, SDXL, and SD1.5 in ComfyUI by 4GB, 6GB, 8GB, and 12GB VRAM limits, SDXL 8GB settings, Flux VRAM requirements, ControlNet compatibility, and CUDA out of memory risk.

Beginner9 minUpdated 2026-05-18
English annotated ComfyUI first image workflow with Queue Prompt and output preview
annotated English visual prepared for this guide family.
Reliable tutorial format: each page starts with the fastest working path, then adds checkpoints, common mistakes, success criteria, and related next steps.

Quick answer

ComfyUI Flux vs SDXL vs SD1.5: VRAM Model Choice should be tested with a small, known-good workflow before you add extra models, nodes, or high-resolution settings.

Recommended workflow

  1. 01

    ComfyUI Flux vs SDXL vs SD1.5: VRAM Model Choice should be tested with a small, known-good workflow before you add extra models, nodes, or high-resolution settings.

  2. 02

    Identify the model type before moving the file. A .safetensors file can still be a checkpoint, LoRA, VAE, ControlNet, or another model type.

  3. 03

    Match each model to the correct ComfyUI folder and to the same base model family used by the workflow.

  4. 04

    Restart ComfyUI or refresh the model list after moving files, then select an installed local model from the node dropdown.

  5. 05

    If a workflow names a file you do not have, choose a compatible local equivalent instead of renaming random files blindly.

  6. 06

    Keep a clean recovery point so you can return to the last working setup if a plugin, model, or dependency change breaks ComfyUI.

Check before you run

  • Use SD1.5 as the first learning model on 4GB, 6GB, and most 8GB beginner setups.
  • Use SDXL on 8GB only with batch size 1, conservative resolution, and no heavy branches in the first pass.
  • Treat Flux as a high-VRAM or cloud-first option unless you already understand ComfyUI model families and workflow dependencies.

Common mistakes

  • Starting with Flux before learning checkpoint, LoRA, VAE, and workflow import basics.
  • Assuming SDXL or Flux files can replace SD1.5 files in any workflow.
  • Debugging prompts when the real problem is VRAM pressure or model-family mismatch.

Success standard

  • You know which family fits your GPU before downloading large model files.
  • You can explain why 8GB SDXL needs a smaller baseline than SD1.5.
  • You know when to use the low VRAM, CUDA OOM, and SD1.5 vs SDXL guides next.

What to do next

  • Open the SD1.5 vs SDXL guide for the detailed 8GB SDXL branch.
  • Open the 8GB VRAM settings guide for exact batch, size, and upscale limits.
  • Open the CUDA out of memory guide if the selected workflow fails during Queue.

Need more context?

This English guide gives the direct working path first. The paired Chinese reference can provide extra screenshots, local download notes, and longer troubleshooting branches for the same topic.

Open the paired reference