ComfyUI: Fast image generation with Flux Schnell

ComfyUI Try it →Watch full video →Artificial Images · Aug 2024

Demo summary

The user switches to the Flux Schnell model and reduces the sampling steps to four to demonstrate high-speed image generation.

Step-by-step

Remove any existing batching or denoising nodes
Add an Empty Latent Image node
Load the Flux Schnell model
Set the sampling steps to 4
Leave the weighting types at default
Click Queue Prompt to generate the images

Options

Adjust weighting types
Increase batch size to generate multiple images simultaneously

Watch out for

The Schnell model is approximately 18GB and takes time to load initially
Image quality is slightly lower than the Flux Dev model

Tips

Use the Schnell model for rapid idea exploration or video-like generation
Expect faster performance after the model is cached following the first run
Be careful not to double-click the prompt button as it may restart the generation process

Highlights

“this model works pretty well it's pretty cool”

All demos from “Getting Started with Flux in RunComfy”

0:180:54Launching a Flux workflow in RunComfyThe user selects the 'comfy y flux' workflow from the RunComfy dashboard and configures the GPU instance size to launch the environment.RunComfy· AI Image Generator
1:313:07Configuring Flux nodes in ComfyUIA walkthrough of the Flux-specific nodes in ComfyUI, including the dual clip loaders, the basic guider, and the sampler custom advanced node used for Flux generation.ComfyUI· AI Image Generator
7:020:23Text-to-image generation with FluxThe user demonstrates Flux's prompt adherence by generating an image of an old TV with the word 'flux' on it in an abandoned workshop.ComfyUI· Text to Image
7:252:04Testing prompt adherence in FluxA demonstration of complex prompt adherence by generating a high-quality image of a dog riding a dolphin at sunset.RunComfy· Text to Image
9:314:03Image-to-image denoising with FluxThe user demonstrates how Flux handles image-to-image tasks by adjusting the denoise strength from 0.4 to 1.0 to show how it differs from Stable Diffusion.ComfyUI· Image to Image
13:342:35Fast image generation with Flux SchnellCurrentThe user switches to the Flux Schnell model and reduces the sampling steps to four to demonstrate high-speed image generation.ComfyUI· AI Image Generator
Watch “Getting Started with Flux in RunComfy” →

AI Image Generator

8:418:40Building a ComfyUI workflow from scratchA step-by-step walkthrough of adding nodes (Checkpoint, CLIP Text Encode, KSampler, VAE Decode) and connecting them to build a custom image generation pipeline.AI Search
19:571:09Generating a medieval warrior with batch processingDemonstrates generating multiple images of a medieval warrior by increasing the batch size in the Empty Latent Image node.AI Search
25:330:32Importing workflows from image metadataShows how to drag an existing AI-generated image into the ComfyUI canvas to automatically reconstruct the entire node workflow used to create it.AI Search
58:094:05Controlling character poses with ControlNetDemonstrates using the OpenPose pre-processor and ControlNet Union model to force a generated princess character into a specific pose from a reference image.AI Search
1:360:57Install custom workflows and missing nodes in ComfyUIThe video demonstrates how to drag and drop a JSON workflow file into ComfyUI and use the Manager to identify and install missing custom nodes.Sebastian Kamph
3:020:20Search and install models via Model ManagerA walkthrough of using the ComfyUI Model Manager to search for specific AI models, such as ControlNet, and install them directly.Sebastian Kamph
3:281:10Running and managing image generation queuesDemonstration of starting a workflow using the 'Run' button, setting up 'Run Instant' for continuous generation, and using 'Run on Change' to trigger generation upon prompt edits.Sebastian Kamph
4:590:55Navigating the ComfyUI workspace and UI settingsShows how to use zoom, fit view, pan modes, and toggle link visibility or themes (light/dark mode) within the ComfyUI interface.Sebastian Kamph
11:311:49Adding and connecting nodes in ComfyUIThe video demonstrates adding a 'Load Image' node via the library or right-click menu and connecting it to other nodes like 'Save Image' using spaghetti links.Sebastian Kamph
19:431:34Converting widgets to inputs in ComfyUIShows how to convert a node's internal settings (like seed) into an external input pin and connect a 'Primitive' node to control it.Sebastian Kamph
25:032:06Managing seeds and randomizationShows the difference between 'Randomize' and 'Fixed' seed settings in the KSampler to either generate new variations or reproduce the same image.Sebastian Kamph
36:450:34Saving images to specific foldersShows how to use the 'Save Image' node and file path syntax to organize output images into custom subfolders.Sebastian Kamph