Flux 3 Klein Character Dataset Creator
Comfy Manager is SLOW to upload. sometimes a day or so. If you get ANY missing nodes, just use the repos listed below.
LC123 Nodes: https://github.com/lonecatone23/ComfyUI_LC123_nodes
LC Vision Nodes: https://github.com/lonecatone23/ComfyUI_LC_Vision_nodes
LC AV Nodes: https://github.com/lonecatone23/ComfyUI_LC_AV_nodes
LC Maskmaker Nodes: https://github.com/lonecatone23/ComfyUI_LC_MaskMaker_nodes
All LC Notes now translate to the following:
English, 中文, 繁體中文, Русский, 日本語 ,한국어, Français, Español, عربي, Türkçe, Português (BR), فارسی, עברית, Italiano
Instagram: https://www.instagram.com/synth.studio.models/
Buy me a☕ https://ko-fi.com/lonecatone
This represents many 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 😉
There's nothing better out there for making character datasets. Not Qwen 2.1. Nothing
V1.0
Flux 3 Klein is exceptional at image edit, so I rebuilt my Krea2 Dataset Creator around it. No ID Edit LoRA needed. Two pictures in, a full dataset out.
✋ Flux 3 Klein is gated. Accept the FLUX Kommunity License on Hugging Face first: https://huggingface.co/black-forest-labs/flux-3-image-klein-comfy
Needs ComfyUI 0.41 or newer and the latest LC123 and LC Vision nodes
Views that actually turn
Flux 3 Klein ignores "profile view". It follows exact wording: "true 90-degree side profile facing the right edge of the frame, with only one eye visible"
Every prompt in all 8 lists is rewritten that way: framing, camera height, pose, where the face points, outfit, background and light
Real side profiles and real rear views (with the short prompts, Flux 3 Klein gave me the same front portrait four times 🤔)
Character Anchor ⚓
Type a few notes about your character ("blonde hair, fit")
My Prompt Enhancer looks at both pictures and writes the full character description once per list. Face, eyes, hair, skin, build. Never clothing, so every prompt sets its own outfit
Seed is fixed, so the description stays the same run after run
Image 1 is the face, image 2 is the body. Remove background still cleans both before they go in
Model Optimizer ⚡ picks the right Flux 3 Klein file for your card. int8 is the fastest and looks almost identical to bf16
About 20 seconds an image on my 5090
Speed Boost🚀 is off. I have yet to fully test it
Everything from the Krea2 version carries over: Vision Assistant with the guided tour, post processing, preview windows and the save logic that keeps your files numbered with their dataset