Open editor →

Mix the references

Compose one picture from separate who / wear / hold / where references.

All four references land in one coherent shot · 2026-10-16 · Reported saved run: $0.053. meta/muse-image/text-to-image · blackforestlabs/flux-3/klein/edit-image

Show multi-reference compositing: one Edit node takes several images in port order, and the instruction addresses each by number (image 1 is the character, image 2 the outfit…), so each part comes from its own source instead of a long prompt.

Saved result

Image 1 · character: Muse 1:1 · reported $0.011 — Mix the references
Image 1 · character: Muse 1:1 · reported $0.011 · Download
Image 2 · outfit: Muse 1:1 · reported $0.011 — Mix the references
Image 2 · outfit: Muse 1:1 · reported $0.011 · Download
Image 3 · prop: Muse 1:1 · reported $0.011 — Mix the references
Image 3 · prop: Muse 1:1 · reported $0.011 · Download
Image 4 · place: Muse 1:1 · reported $0.011 — Mix the references
Image 4 · place: Muse 1:1 · reported $0.011 · Download
The mix: FLUX.3 Klein Edit, 4 references, 1024×1024 · reported $0.009 — Mix the references
The mix: FLUX.3 Klein Edit, 4 references, 1024×1024 · reported $0.009 · Download

Four Muse stills stand in for your own uploads: a paper-craft fox (image 1), a yellow knitted raincoat (image 2), a red paper lantern (image 3) and a paper harbor pier at dusk (image 4). FLUX.3 Klein Edit gets them in port order (image, image2, image3, image4), and the mix instruction names each one by number. In the result the fox wears the same raincoat with its three wooden toggles and holds the one lantern by its bamboo handle while standing on the cardboard pier, with the striped lighthouse behind it. Flaws: the two paper boats from image 4 mostly drop out, the fox's ears poke through the hood, and the coat comes out longer than the flat-lay reference. A first prompt without “exactly one lantern” gave the fox two lanterns, so the instruction now says it outright. Swap any reference Image node for a 📷 Image input to mix your own photos. Checked outputs are SFW (no people). Inspired by the multi-reference workflows ComfyUI showed for Krea 3 and FLUX.3 Klein (https://x.com/ComfyUI/status/2110840667337642330); no media from that post is used.

Make it yours

  1. Open the graph — save first if the canvas already has your work. Undo brings yours back.
  2. Each reference has its own prompt: Who · character, What they wear · outfit, What they hold · prop and Where · place. Rewrite any of them, or replace its Image node with a 📷 Image input and drop in your own photo (a product, a mascot, a jacket, a room). Keep the wiring order and update the matching “Image N is…” line in Mix instruction (names each image). Klein takes up to 10 references: wire image5 and name it the same way.
  3. Check the editor estimate, then run with your NanoGPT key. Four Muse 1:1 stills at $0.011 each and one FLUX.3 Klein Edit 1024×1024 at $0.009. The saved run was $0.053. With your own uploads instead of the Muse stills it is about $0.01.
  4. Check each reference against the mix: same fox colors and face, same coat and toggles, exactly one lantern, same pier and lighthouse. Look for parts that drop out (here the paper boats) and parts that change shape (the coat got longer).
Open this noodle → See the reviewed run

Download workflow · Original sampled graph · Explore MCP tools · awesome-noodles