🐰Flux.1 Kontext - Editing (Multi Image)

This is the fusion node: feed it two, three, or four separate reference images and it hands back one image that plausibly contains all of them together. The pack's own demo is the clean version of the use case - a photo of a person and a separate photo of a dog, composited into one scene of them sitting on the same couch. That's meaningfully harder than single-image editing, because the model isn't just changing one thing, it has to reconcile lighting, scale, and pose logic across subjects that were never photographed together.

Mechanically it's the same in-context trick as the rest of the pack, just with more references stuffed into the context window at once - up to four images alongside your prompt, reasoned over together rather than composited after the fact the way IP-Adapter face/style stacking or a ControlNet pass would do it. Worth calibrating your expectations here, because this is the one place people report Kontext falling short of what they hoped for: users trying to use multi-image input for pose transfer - take the pose from image two and apply it to the character in image one - report getting back something close to the unchanged first image, because that's not actually what this mode does. It's a single-pass fusion of subjects into one coherent scene, not a control mechanism for lifting a specific attribute (pose, outfit, lighting) from one image onto another. If that's what you're after, this node will disappoint you; if you want "these things, together, in one shot," it's the right tool.

The four image inputs - image_1 through image_4 - are all optional at the schema level, which means ComfyUI won't stop you from running the node with one or even zero connected. Don't. Zero images makes this the text-to-image node with extra steps, and one image makes it the single-image editing node with extra steps; the fusion behavior needs at least two to have anything to reconcile. Write your prompt around the relationship, not each image individually - "the woman from the first photo sitting next to the dog from the second, on a park bench" rather than two separate descriptions - since describing the composition is what actually drives how Kontext fuses the references. Everything else - model (pro vs. max, Max being the pricier no-compromise tier), num_images, seed, guidance_scale (3.5 default), num_inference_steps (28), aspect_ratio, output_format, safety_tolerance, prompt_upsampling - works identically to the other two nodes in the pack. One thing the README doesn't call out: pricing is quoted per model tier (Pro or Max), not per reference image, so feeding it four images costs the same as feeding it two by the pack's own numbers. Worth a sanity check against your TuZi bill if you're running this at volume, since that's the kind of detail resellers sometimes change without updating the docs.

Outputs are image and status, same as the rest of the pack - status is worth watching here specifically, because a multi-image call has more ways to fail (a dropped connection on one of four image slots, a payload that's too large) and the status string is where that shows up instead of a mysterious blank result.

The full-body anatomy limitation that affects the single-image editing node gets worse here: you're now fitting multiple people into one frame, which pushes each of them below the size threshold where Kontext's heads and hair start artifacting. If you're fusing two full-body subjects into one scene and the faces come out mushy, that's the same known limitation, just compounded - crop tighter, favor upper-body/portrait framing over full-body group shots, or refine the result locally afterward.

Install is identical to the rest of the pack: ComfyUI Manager → "Install via Git URL" with https://github.com/LKbaba/ComfyUI-TuZi-Flux-Kontext.git, or git clone it into custom_nodes/ and pip install -r requirements.txt (requests, python-dotenv, fal-client, httpx, httpcore; portable installs run that through python_embeded\python.exe). Restart once to generate the .env file, close ComfyUI, put your TUZI_API_KEY into ComfyUI/custom_nodes/ComfyUI-TuZi-Flux-Kontext/.env (get a key at api.tu-zi.com/panel), restart again.

Troubleshooting follows the same pattern as the pack's other nodes: a red node on load means .env is missing or the key's a placeholder; a node that loads but won't generate means check your TuZi balance and key validity before suspecting the node itself; missing entirely after install means confirm fal-client installed and do a full restart. It's also worth knowing before you commit to a workflow built around this specific node that these third-party API-relay nodes draw real skepticism in the ComfyUI community - people who've dug into similar Kontext relay nodes describe them as "just a shell" around someone else's paid endpoint, which is accurate and not a defect, just something to know you're buying before you wire up a complex fusion workflow around it.