TTP Tile Preprocessor Simple

TTPlanetPig isn't new to ControlNet Tile - before this pack, they released TTPLanet_SDXL_Controlnet_Tile_Realistic, a well-regarded SDXL tile ControlNet checkpoint that got a genuinely positive Reddit reception ("this one is awesome," a few dozen upvotes, some debate over whether it beats the SD1.5 original - the honest community verdict was "it's really hard to make good CN-tile-XL models," not that this one was bad). This node is that same lineage's preprocessor: it produces the actual conditioning image a Tile ControlNet expects, so you don't have to build it by hand.

What Tile preprocessing actually does, and why it exists. The tile ControlNet condition works by feeding the model a version of the original image with the fine detail stripped out but the overall structure and color intact. That gives the sampler two things at once: freedom to invent new detail (because the low-frequency guide doesn't specify it), and a leash that keeps the invented detail faithful to the source composition. This is, in the community's own words, "the foundation of tiled upscaling workflows" - pair it with tiled diffusion and you get one of the most reliable photorealistic upscale recipes going.

How this node does it. Two knobs, both genuinely load-bearing: scale_factor (1–8, default 2, step 0.05) downsamples the image before processing - a factor of 2 means you're throwing away roughly half the linear detail before the blur pass, which is what forces the sampler to actually generate new fine detail instead of just copying what's already there. blur_strength (1–20, default 1, step 0.1) then softens what's left. Push blur_strength up and you get a softer, more permissive guide (the model has more room to hallucinate, at the cost of faithfulness); keep it low and the tile stays closer to the source. Output is a single image_output - wire it straight into your ControlNet's "apply" node alongside the Tile model.

Wiring it into a real workflow. image → TTPlanet_Tile_Preprocessor_Simple → ControlNet Apply (Tile model) → your img2img/upscale KSampler. Nothing fancier than that. If you're doing a full tiled upscale, this typically sits per-tile, right after TTP_Image_Tile_Batch slices your source and before each tile goes through the sampler.

Install. ComfyUI Manager, search "Comfyui_TTP_Toolset", or cd ComfyUI/custom_nodes && git clone https://github.com/TTPlanetPig/Comfyui_TTP_Toolset, then restart. The preprocessor itself needs no extra downloads - but it's useless on its own. You still need an actual Tile ControlNet model checkpoint that matches your base checkpoint's architecture (SD1.5, SDXL, or a newer family's tile model if one exists), loaded separately through your usual ControlNet loader.

Common issues. The single most common way people wreck a tiled upscale is skipping the ControlNet Tile step entirely and just sampling raw tiles - without it, adjacent tiles have nothing forcing them to agree with each other, and you get visible patchwork seams where they're stitched back together. If you're already using this preprocessor and still seeing seams, check two things: that your ControlNet Tile model actually matches your base checkpoint's architecture (a mismatched tile model is a silent failure, not an error), and that your blur_strength isn't so high the conditioning has lost the structural information that was supposed to keep tiles aligned in the first place. And architecture matters more than it looks like it should here - an SDXL tile model on an SD1.5 checkpoint (or vice versa) won't error out cleanly, it'll just produce mediocre, slightly-off results that are easy to blame on the wrong node.