Photoshoot Schwarzweiß

Ever rendered what you thought was a monochrome shot and the lips came back with a warm brown that no amount of "black and white photograph, monochrome" in the prompt would argue out of the model? That's not you doing it wrong. That's Krea 2 - or any distilled, CFG-1 model - running out of patience near the bottom of the colour channel spread. Krea2Monochrome (shown on the canvas as Photoshoot Schwarzweiß) is the one node in the ComfyUI-Photoshoot kit that touches pixels instead of words, and it exists to kill that last trace of colour after the sampler has done its thing.

What it actually does

The Photoshoot philosophy is "labels in, English prompt out" - everything else in the pack just writes text your sampler reads. Colour, however, only exists after VAE Decode, which is why the Style node can't be the whole answer: it runs before the sampler and hands out a string. So this node works in image space, hung between VAE Decode and Save Image.

Mechanically it's refreshingly unmagical, about 15 lines of torch. It converts via Rec. 709 luma weights - (0.2126, 0.7152, 0.0722) - not a naive channel average. That matters: equal weights would map red lips and green foliage to the same grey and flatten your image into soup. Proper luma keeps the tonal separation a real monochrome conversion gives you. Then it optionally tints the grey and blends the original RGB toward the tinted result by the strength slider. At the default strength of 1.0 you get true, tinted greyscale. Anything less is just a desaturated colour image - which the author himself notes is "sometimes the point."

The inputs that matter

Only three, and only two you'll ever touch:

• images (IMAGE) - whatever came out of VAE Decode. Batch-friendly: drop a whole series or a stack of frames through and every one gets converted. Extra channels survive, so you're not wrecking an alpha mask.
• toenung - the tint, and yes, that's German for "tint." ComfyUI's locale system translates input names but not combo values, so the choices are words both languages read the same: Neutral, Sepia (warm), or Selenium (cool). The tints are deliberately mild - a sepia that's obviously brown is just a colour image again.
• staerke (strength, default 1.0) - leave it at 1 unless you're hunting a specific desaturated look.

Output is images, straight into Save Image.

Wiring it

Decode → this node → Save Image, then select the node and hit Ctrl+B. That's it. The pack's example workflows already carry it in place, shipped bypassed, because most people shooting colour don't want it on. Ctrl+B flips it on; the prompt still decides the tonality of the result, so pair it with a black-and-white look from the Style node. Run a full-colour image through it and you get a colour photo that was converted, not one that was shot in black and white - the node removes colour, it can't invent the mood.

Installing

Same as the rest of the pack, no model downloads, no dependencies beyond ComfyUI itself (the project's pyproject.toml declares an empty dependency list). ComfyUI Manager: search for Photoshoot. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/ralksta/ComfyUI-Photoshoot

Restart ComfyUI and it appears under the Photoshoot category. The interface follows ComfyUI's locale setting - a German browser gives you German labels with no action on your part; pick a language under Settings → Comfy → Locale if that surprises you.

Where people get burned

The big one: don't expect this node to rescue a colour render into something that feels monochrome. It's a finisher, not a substitute for the prompt - the author's own measurements found lipstick keeping up to 21 units of channel spread out of ~25 in a "black and white" render. This node closes exactly that gap, nothing more. If your output still looks tinted after running it, check staerke - a value below 1 is deliberately partial. And if you're feeding it a LATENT because you're used to working pre-decode, it won't accept it: it needs the decoded IMAGE.

It's a young, niche node from a young pack - comfy.icu shows zero impressions at the time of writing. But it solves a real, measured annoyance that no amount of prompt engineering fixes. If you shoot a lot of monochrome series, this is the two-minute addition worth having.