All-in-One FluxDev v0.2

Summary

Large multi-purpose FLUX.1-dev workflow that combines text-to-image, image-to-image, LoRA loading, ControlNet, negative-prompt-style experimentation, simple inpainting, and iterative upscale/detailing in one graph.

Model Family

flux

Status

imported-only

Source

Original page: https://github.com/Ling-APE/ComfyUI-All-in-One-FluxDev-Workflow
Workflow file: https://raw.githubusercontent.com/Ling-APE/ComfyUI-All-in-One-FluxDev-Workflow/main/All-in-One-FluxDev-v0.2.json
Original author: Ling-APE
Date imported: 2026-08-08

Developer notes

The author describes this as a personal "all in one" interpretation rather than the single correct way to use FLUX.
The workflow leans heavily on Use Everywhere nodes to keep a large graph manageable.
It supports switching between text-to-image and image-to-image, optional ControlNet and LoRA loading, two different sampling approaches, simple inpainting, and a high-res fix style upscale path.
The author notes that the included XLabs ControlNet tests worked best with minimal prompting, square-ish setups, and guidance around 4.
A converted ComfyUI-friendly version of the XLabs realism LoRA is recommended by the author.
The repo also mentions optional Ollama + Florence2 helpers for captioning and prompt assistance.

Our notes

This is the most feature-rich import in the current flux folder, but it is also the messiest dependency-wise.
Good candidate when you want one workflow to dissect and mine for ideas rather than the cleanest daily-driver graph.
If you only want one FLUX workflow that covers a lot of territory, this is the most "toolbox" option of the five.
Expect node drift over time because it pulls from several custom packs and a 2024-era FLUX experimentation stack.

Required custom nodes

cg-use-everywhere / Use Everywhere
ComfyUI-Inspire-Pack
ComfyUI-Impact-Pack
ComfyUI-TiledDiffusion
ComfyUI-Florence2
comfyui-ollama
rgthree-comfy
pysssss custom scripts
ComfyUI-Crystools
ComfyUI-PixelResolutionCalculator
likely ComfyUI-KJNodes or equivalent schedule-hook providers used by the upscale section

Built-in / core nodes used

FLUX guidance and sampling nodes
UNET / VAE / CLIP loaders
ControlNet loader/apply
LoRA loader
VAE encode/decode

Notes

Best use: hybrid experimentation, not a minimal clean import.
If missing-node resolution gets annoying, start with the simpler workflows in this repo first and come back to this one later.