Скачать или смотреть Mastering Flux: Full Workflow With Flux 1, Flux 2, Krea & Kontext (All-In-One Tutorial)

Welcome back to Arctic Latent!
In this massive in-depth tutorial, we build a complete all-in-one Flux workflow using Flux 1, Flux 2, Krea, and Kontext, covering everything from the basics to advanced creative pipelines inside ComfyUI.

This is the most comprehensive Flux-based workflow I’ve made so far — we build text-to-image, image-to-image, upscaling, inpainting, outpainting, character consistency, pose transfer, face swapping, and much more, all inside one unified setup.

00:00:00 Intro
00:01:45 LinuxComfyUI Installer
00:04:07 Installing Custom Nodes
00:06:48 Global Settings
00:10:55 Text to Image
00:16:09 Image to Image
00:19:38 Advanced Upscale
00:23:05 Simple Upscale
00:24:05 Face Detailer
00:28:46 Inpainting
00:38:06 Outpainting
00:40:32 Depth
00:45:33 Canny Edges
00:48:55 PULID
00:54:19 Swap Faces
00:55:30 Flux Krea
00:57:50 Kontext
01:02:43 Kontext Image Chain
01:05:31 Kontext Inpainting
01:08:24 Flux 2
01:16:17 Attention Logic
01:19:22 Benchmarking Flux 2
01:23:59 Pose to Image
01:27:49 Fancy Controls
01:31:27 Add and Use LoRAs
01:36:52 Outro


We also dive into performance benchmarking, comparing:

Native PyTorch Attention

Sage Attention

Sage Attention with FP16 Accumulation

You’ll see real-world generation times, stability differences, and how FP16 accumulation changes performance.

🔥 What You’ll Learn in This Video

This tutorial includes every major workflow type, all wired together:

✨ Core Image Workflows

Text-to-Image (Flux 1 & Flux 2)

Image-to-Image refinements

Advanced Upscaling (multi-step)

Simple Upscaling (fast mode)

Face Detailer for portrait enhancement

Pulid for consistent character design

Flux 2 with reference images for character/style alignment

🎨 Editing Workflows

Inpainting (Flux + Kontext)

Outpainting (scene expansion)

Kontext Inpainting & Rewriting

Kontext Image Chaining for multi-stage editing

Basic Kontext Editing for iterative prompt refinement

🧩 Structure & Control

Depth to Image workflow

Canny Edge to Image workflow

Pose-to-Image (pose-guided generation)

Face Swapping integrated into the pipeline

⚡ Performance & Optimization

How Sage Attention works

Why FP16 accumulation matters

Benchmarking native attention vs Sage Attention

Speed differences between Flux 1 and Flux 2

GPU optimization tips for faster generation

📊 Benchmarking Breakdown

We test the same prompt across three systems:

Native PyTorch Attention

SageAttention

SageAttention + FP16 Accumulation

You’ll see:

Per-step speed

Total generation time

Memory usage

Any visible quality differences

🧊 Tools & Models Used

Flux 1 Dev + Flux 1 Krea

Flux 2

Kontext (editing + chaining + inpainting)

Krea prompt enhancements

SageAttention 2.x

Pulid for character consistency

Face Detailer, Outpainting Nodes, ControlNet edges/depth

🙌 Support Arctic Latent

If these deep-dive tutorials help you, please like and subscribe — it truly motivates me to keep making bigger, better videos.

💙 Patreon:
  / arcticlatent  

🤖 HuggingFace:
https://huggingface.co/arcticlatent

💻 GitHub:
https://github.com/arcticlatent

⚠️ Disclaimer:
All images shown in this video are generated using the Flux models developed by Black Forest Labs. I am not the original artist or rights holder of the training data used to build these models. The visuals in this video are solely the output of the Flux model and are used for educational and demonstration purposes only.