🎛️ Smart USDU Universal (No Upscale)

Here's where the pack's architecture becomes visible. The Smart Tile Solver V6.2 computes an optimal tile grid and hands you a bundle of numbers; this node - Smart USDU Universal (No Upscale) - is built to receive those numbers. It drops the internal upscaling and takes the pre-upscaled upscaled_image plus explicit output_width, output_height, tiles_x, tiles_y, tile_width, tile_height, and tile_padding inputs, then runs the full Universal pipeline (diffdiff + ControlNet toggles) on exactly that grid.

The design intent is a clean split: the solver plans and upscales, this node does the tiled diffusion pass. No re-deriving grid math inside the node, no drift between what was planned and what gets sampled.

The important bit: safe_guard

safe_guard defaults to True and validates that the values match what the solver produced - the source throws an error on any mismatch. That's a real convenience when you're wiring the solver's INT outputs straight across: it catches a mis-typed tile_width before it wastes a whole render. If you're deliberately overriding the solver's plan, flip it off - but then you own the consequences.

Otherwise it's the Universal feature set: enable_diffdiff / enable_controlnet toggles, per-tile conditionings, the sampler block, mode_type, mask_blur, seam-fix settings, tiled_decode. Optional block: denoise_mask, multiplier, model_patch, control_image, control_strength, control_mask. Note there's no upscale_model or upscale_by anywhere - the upscale already happened. Output is a single IMAGE.

Installing it

Part of ComfyUI-ArchAi3d-Qwen by Amir Ferdos (ArchAi3d). ComfyUI Manager → search "ArchAi3d Qwen" → install → restart, or:

cd ComfyUI/custom_nodes
git clone https://github.com/amir84ferdos/ComfyUI-ArchAi3d-Qwen.git
cd ComfyUI-ArchAi3d-Qwen
pip install -r requirements.txt

Restart after. Only ComfyUI's own torch/numpy/Pillow are needed; the heavyweight deps in requirements.txt belong to other pack nodes. Free for personal use; commercial use requires the paid license.

Common gotchas

The #1 failure is feeding the raw source image in - everything downstream then runs at 1.0 scale with a grid that doesn't match. Keep safe_guard on while you build the graph so a mismatch errors loudly instead of silently producing garbage. And remember tile_width/tile_height here are the solver's output-space tile sizes, not your input image's dimensions.