TTPlanet Tile Simple

This is the preprocessor for TTPlanet's Tile ControlNet, a popular realistic-tile model line. Tile conditioning is the backbone of detail-preserving upscaling: you feed the original image back to the model as a condition so each tile stays faithful to the source while the model re-adds detail. But here's the counterintuitive part - tile control wants a slightly degraded input. This node does the degrading: it blurs and optionally rescales your image so the Tile ControlNet has room to hallucinate crisp detail back in.

How it works, and why blur is the point

If you hand a Tile ControlNet a perfectly sharp image, there's nothing for it to improve - it just tries to reproduce what's already there. The trick that makes tiled upscaling work is to give it a softened version: the model sees the structure and color faithfully, but the fine detail is missing, so it fills that in at your target resolution. That's what the blur is for. It's not a mistake, it's the mechanism. TTPlanet's tile models are trained expecting this kind of pre-blurred input, which is why they pair with this specific preprocessor.

"Simple" distinguishes it from the guided-filter variant in the pack (TTPlanet_TileGF_Preprocessor). Simple is the lighter, more direct version - blur and go.

The inputs and output

Just two knobs, and both shape the degrade:

• image - your source (typically a tile or the whole image mid-upscale).
• scale_factor (default 1, range 1–8) - how much to rescale before processing. Leave at 1 for a straight blur pass; raise it if your workflow wants the preprocessor doing some of the resizing.
• blur_strength (default 2, range 1–10) - the important one. How much to soften the image. Too little and the Tile ControlNet has nothing to work with (output looks like the input); too much and you lose the structure the tile is supposed to preserve. The default 2 is a reasonable starting point - tune from there based on how much detail you want reinvented.

The single output is an IMAGE - the blurred/degraded map - which wires into a ControlNet Apply node with a TTPlanet (or compatible) Tile ControlNet, inside a tiled-upscale setup.

Installing it

ComfyUI Manager: search ComfyUI's ControlNet Auxiliary Preprocessors, install, restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/Fannovel16/comfyui_controlnet_aux
pip install -r requirements.txt

Restart ComfyUI. This node is image processing (blur/resize), so there's no preprocessor model to download - but you do need to grab a Tile ControlNet model separately for the actual conditioning.

Where people get burned

First and biggest: this node makes the hint, but the tile magic lives in the ControlNet model you pair it with plus the tiled upscaling setup around it. On its own, a blurred image does nothing. The full picture is this preprocessor → a Tile ControlNet → a tiling upscaler like Ultimate SD Upscale or Tiled Diffusion. Wire up only the preprocessor and you'll wonder why nothing sharpens.

Second, blur_strength is a genuine balance and the difference between a great upscale and a smeared one. Under-blur and the model has no room to add detail; over-blur and you've thrown away the structure it was supposed to keep faithful. If your upscale looks either unchanged or hallucinated-wrong, this slider is the first thing to adjust.

Third, match the preprocessor to the model. TTPlanet tile models expect this pre-blur; a different tile ControlNet may want a different input. Read the model card for whichever Tile ControlNet you loaded, and note that tile quality has historically varied a lot across base models - check that your tile model actually suits the checkpoint you're upscaling on.