Save 16-bit TIFF

Here's the quiet punchline of the entire Bit Depth Enhancer pack: its enhancement nodes hand you a float32 tensor with fractional tonal steps, and if you save that with ComfyUI's default Save Image node, it gets rounded straight back into 8-bit banding. All that work, undone by a PNG. Save16BitTIFF is the node that makes the pack's promise real - it writes genuine 16-bit TIFF files with 65,535 levels per channel, the kind of file Photoshop, DaVinci Resolve, and print workflows actually respect.

So the workflow that makes sense of this whole pack is: Load Image → Bit Depth Enhancement (Classical) or ABCD or deepDeband → Save 16-bit TIFF. The enhancement nodes produce the precision; this node commits it to disk. Skip it and you're fighting the point.

How it works

Straight out of the source: the node takes your images tensor, multiplies by 65,535, casts to uint16, and writes it with tifffile using lossless DEFLATE compression straight into ComfyUI's output folder (ComfyUI/output/, via the standard folder_paths lookup). Filenames are {prefix}{timestamp}_{seq}.tif, so batches come out as numbered sequences and you never overwrite a run. It's an output node - no return values, it just files the images and reports them back to the UI.

Two design decisions in the code are worth understanding before you trust them:

It always saves Linear RGB. The color_profile dropdown - sRGB, Adobe RGB, ProPhoto RGB, Linear - is cosmetic. Every profile writes linear data, and the selection only ends up in a UserComment metadata tag. That's deliberate: ComfyUI works in linear space, so applying an sRGB gamma curve at export would double-encode the data. The author's tooltip says it plainly: apply your final color space in Resolve or Photoshop, not here. The trap is expecting the dropdown to convert your image - it won't. If you genuinely need an sRGB-encoded TIFF, convert the image before this node or in your delivery tool.

It's the only honest 16-bit path in the pack. The enhancement nodes output float32 values in [0, 1]; this is what turns that precision into real 16-bit integer levels on disk.

The inputs

There are three, and only one needs your attention:

• images - whatever the enhancement node fed you. Feed it 8-bit banded garbage and it'll happily write 16-bit garbage; precision doesn't fix content.
• filename_prefix (default enhanced_16bit_) - set something meaningful per project. The timestamp + sequence suffix keeps collisions from happening.
• color_profile (sRGB / Adobe RGB / ProPhoto RGB / Linear, default sRGB) - remember: metadata only. Pick whichever you'll note down in your delivery pipeline.

Installing it

It's the same Bit Depth Enhancer install as the rest of the pack. ComfyUI Manager (search "Bit Depth Enhancer") or:

cd ComfyUI/custom_nodes
git clone https://github.com/subraoul/ComfyUI_Bit-Depth-Enhancer.git
cd ComfyUI_Bit-Depth-Enhancer
pip install -e .

Restart ComfyUI. The one dependency that matters here is tifffile, which the pack's pyproject pulls in automatically (and the code refuses gracefully if it's missing - it just prints an error and returns nothing, so check the console if you get no files).

Common issues

• No files appear - the node writes to ComfyUI's output directory, not next to your workflow. Look in ComfyUI/output/. If tifffile is missing, you'll see the error in the console instead.
• "My TIFF looks washed out in Photoshop" - that's linear data displayed without a color profile, not a bug. Assign your working profile (or let Resolve's color management handle it) and it'll look right.
• The color profile dropdown "does nothing" - correct, by design. It's a note-to-self in the metadata. Don't fight it; just convert downstream.
• Big 4K batches fill the disk - 16-bit linear TIFFs are chunky, and DEFLATE only does so much. Clear output/ between projects.

It's a one-job node, and it does that one job properly. If you're taking this pack's output anywhere serious, you need it.