How to Generate Fast Images with Flux.1 Schnell in ComfyUI

Set up the Apache-licensed Flux Schnell model for 4-step generation that runs in seconds, perfect for quick drafts and lower-end cards.

7 minIntermediate

Visit FLUX

Flux.1 Schnell is the distilled, speed-focused sibling of Flux Dev. It is tuned to produce a good image in roughly 4 steps instead of 20-plus, and it ships under the permissive Apache 2.0 license. If you want fast drafts or you are on a weaker GPU, Schnell is the model to reach for.

What you need

A recent ComfyUI install
The same CLIP and VAE files used for Flux Dev (clip_l, t5xxl, ae)
An 8 GB GPU works well; the fp8 build runs on less
About 24 GB of free disk space

Step 1: Download the Schnell model file

Download flux1-schnell.safetensors and place it in models/unet. Schnell shares the text encoders and VAE with Dev, so if you already set up Flux Dev you only need this one new file.

Step 2: Reuse the Flux loaders

Open your Flux workflow and switch the UNET loader from flux1-dev to flux1-schnell. Leave the DualCLIPLoader and VAE loader exactly as they were. Nothing else about the model wiring changes.

ComfyUI - switch the model

Load Diffusion Model

unet: [ flux1-schnell.safetensors v ]

weight_dtype: default

DualCLIPLoader -> unchanged

Load VAE -> unchanged

Only the unet dropdown changes; encoders and VAE stay put.

Step 3: Dial the sampler down to 4 steps

This is the key difference from Dev. Schnell is distilled for very few steps and does not use real CFG, so set steps to 4 and CFG to 1. Use the euler sampler with the simple scheduler. Higher step counts waste time without improving the image.

ComfyUI - KSampler settings

KSampler

----------------------------------------

steps: 4

cfg: 1.0

sampler_name: euler

scheduler: simple

denoise: 1.00

Guidance is baked in

Because Schnell ignores CFG, the FluxGuidance node has little effect here. If your template has one, you can leave it at default and focus on getting the prompt right instead.

Step 4: Queue and compare

Enter a prompt and click Queue Prompt. On a mid-range card the image lands in a few seconds. Run the same prompt a couple of times with different seeds to find a composition you like, then refine the wording.

ComfyUI console - a Schnell run

Requested to load Flux

100%|####################| 4/4 [00:02<00:00, 1.71it/s]

Prompt executed in 3.4 seconds

Result: near-instant Flux renders for ideation. When a draft looks promising, you can rerun the same prompt and seed through Flux Dev at higher steps for a more polished final.