Comfy sometimes takes a day or two to update. I am not missing nodes. You can get all of my node packs here for free if you cannot get them from the ComfyUI Manager: https://github.com/lonecatone23

Flux 3 Klein: what is it?

Flux 3 Klein is Black Forest Labs' new open model. It's a 5.6B image model with Qwen3-VL-4B as its text encoder, and it does three jobs in one: text to image, editing a picture, and building a new picture out of up to 10 reference images. It also takes bounding boxes: draw a box, say what goes in it, and it puts it there. It can write text in the picture, too.

There are two versions. Klein (the files with _sd_ in the name) is distilled: 16 steps at CFG 1, and it's the one you want for making pictures. Base is the undistilled model (50 steps, CFG 3.5), made for training LoRAs. The int8 files run on 8 GB cards, and they're what this workflow uses.

🛑🚫✋ Accept the license first

• The files are gated. Open the [FLUX 3 Klein page](https://huggingface.co/black-forest-labs/flux-3-image-klein-comfy), sign in to Hugging Face and click Agree and access repository, or nothing downloads.

• Every model link, plus how to log ComfyUI in to Hugging Face, is in the Start Here note in the workflow.

• Needs ComfyUI 0.41.0 or newer.

• 💡 The license is non-commercial, plus commercial use of your pictures if you (and your company) make under $5M a year.

Prompting Flux 3

Flux 3 was trained on long, full descriptions of the picture. A short prompt sent as typed gives you generic results (Black Forest Labs says it straight up: don't run it without a prompt upsampler). That's why the LC Vision Prompt Enhancer is on by default and uses Black Forest Labs' own upsampler instructions. Write your rough idea, it writes the full description.

Editing

• Say what changes, then what stays: "Make her parka red. Keep her face, hair, pose and the street exactly the same."

• ⚠️ Describe the change, not the whole scene. Describing everything again redraws it with a new pose and framing.

• The Prompt Enhancer looks at your picture while it writes, so it knows what you're editing.

• Extra reference pictures are ref_image_1 to ref_image_4. Say what you want from each one: "the jacket from ref_image_1".

Text to image

• Turn off Load Image and switch the Prompt Enhancer to Flux 3 Text to Image.

• Framing, subject, background and lighting all help. For a style, describe it (palette, brush strokes, texture) instead of naming an artist or a painting.

• Text goes in quotes: a sign that reads "OPEN".

Boxes 🔲

• Turn on the Boxes group, drag a box on the picture and name it (hat, cup).

• Pick what happens to it: new, change, move or remove.

• Write <1>, <2> in the prompt for box 1, box 2.

• 💡 Boxes guide where things go. They're not a hard mask, so something can spill a little past its box.

What's included

• LC Model Optimizer ⚡: picks the right Flux 3 files for your card (int8, nvfp4, bf16 or GGUF) and turns on the speed-ups that passed the test on your machine.

• Edit with up to 5 reference pictures: image 1 is the one you edit, 2 to 5 are extras you switch on.

• LC Vision Prompt Enhancer 📝: Black Forest Labs' upsampler instructions, and it sees your picture.

• Image Prompt Assist: describes your picture for you.

• Boxes 🔲 (new): LC Bounding Boxes and LC Flux 3 Layout Prompt, my new nodes for placing and editing things by box. They work in the classic canvas and in Nodes 2.0.

• Detailing suite for the face and eyes (SAM 3.1).

• VOSR2 Upscale 🧩: fast, low VRAM, keeps what the detailers fixed.

• Post processing: LC Sharpen Pro, LC Apply LUT, LC Natural Look and LC Image Crop.

• Save with metadata, so CivitAI reads your settings.

• Options switches for everything (no manual bypassing).

• Vision Assistant 🔬 with a guided tour 🗺️.

• All LC Notes translate to English, 中文, 繁體中文, Русский, 日本語, 한국어, Français, Español, عربي, Türkçe, Português (BR), فارسی, עברית and Italiano.

Note: I spend a lot of time on these to make sure they are top notch. If you have ANY issues, I'm very responsive. Hit me up and I'll resolve them for you.

Instagram: https://www.instagram.com/synth.studio.models/

Buy me a ☕ https://ko-fi.com/lonecatone

This represents hundreds of hours of work that I give away for free. If you enjoy it, please make sure to 👍 like and post to my workflows. That is how you support me. Also, feel free to ⚡ tip 😉