📌 Overview

Four reference photos in, one group shot out — locally, no API.

FLUX.2 [dev] run locally with chained ReferenceLatent conditioning: up to four people composed into one picture. No account, no per-image cost.

✅ What you get

• FLUX.2 four-reference workflow

🧰 What you need

• 40 GB+ VRAM

• VRAM: 40 GB+ · Level: Intermediate

📦 Models and packs

Uses the flux2 pack from the Gmanski RunPod template, which installs every checkpoint, LoRA and custom node this workflow needs in one click.

⬇️ Download

Free. The workflow JSON is attached to this listing. Full notes, tested settings and the exact model packs are on its page:

👉 Workflow page on gmanski.com

💬 Help

Questions? Our Discord bot answers most things in under a minute, and a human follows up.

⚠️ Disclaimer

Every face in the samples is a generated character, not a real person. Use responsibly and within the law where you live.

Gmanski makes ComfyUI workflows, a one-click RunPod template, custom nodes and ready-made AI characters: gmanski.com