๐จ Smart Tile Sampler
Sample every tile, leave the stitching to the Merger
- image
- segs
- model
- vae
- conditionings
- negative
- bundle
- image
- segs
In the Smart Tile pipeline there are two ways to process your tiles: the Detailer composites as it goes, and the Sampler takes a cleaner division of labor. This node is the second one - it samples every tile independently and hands the results back as SEGS, doing zero compositing. The Merger downstream does all the stitching. It's a deliberate split: sample everything with your best settings, then decide once how to blend, rather than committing to a blend strategy mid-sampling.
It's part of the "Smart Tile" subsystem of ComfyUI-ArchAi3d-Qwen (Amir Ferdos's pack). The intended workflow is written right in the source: Calculator โ SEGS Blur โ Sampler โ Merger โ Final Image.
Inputs
- image - the upscaled base image. It's a pass-through: the Sampler doesn't modify it, it just carries it along so the Merger can use it as the fallback base.
- segs - the tile SEGS, from Smart Tile SEGS Blur (the blurred-mask version, so the seams are already feathered).
- conditionings / negative - from Smart Tile Conditioning: one conditioning per tile, plus the shared negative.
- model / vae / seed / steps / cfg / sampler_name / scheduler / denoise - the usual sampling stack.
denoisedefaults to 0.4, which is the "refine, don't rebuild" range. - bundle (optional) - from the calculator, carries the geometry.
Outputs: image (the untouched base, passed through) and segs - the SEGS with each tile's rendered result inside it. That segs output is the whole ballgame: it's what the Merger reads to build the final image.
Installing
cd ComfyUI/custom_nodes
git clone https://github.com/amir84ferdos/ComfyUI-ArchAi3d-Qwen.git
cd ComfyUI-ArchAi3d-Qwen
pip install -r requirements.txt
Or ComfyUI Manager โ "ArchAi3d Qwen". No model downloads. Free for personal use; commercial use needs a license.
Sampler vs Detailer
The difference is worth internalizing before you pick. Smart Tile Detailer processes each tile with its own conditioning and composites the result in the same pass - fewer nodes, faster to wire, and it uses SEG masks directly. This node deliberately defers compositing to the Merger, which buys you the Merger's proper normalized weight blending and its four blend modes. If you're chasing maximum seam quality, the Sampler + Merger split is the better architecture; if you just want a detail pass done, the Detailer is less ceremony. Both need the Conditioning node's per-tile prompts to earn their keep - without those, you've built a slower version of a plain tiled upscale.
Inputs (13)
| Name | Type | Default | Description |
|---|---|---|---|
| image | IMAGE | Input image (pass-through to Merger as base) | |
| segs | SEGS | SEGS from Smart Tile SEGS Blur | |
| model | MODEL | Diffusion model for sampling | |
| vae | VAE | VAE for encode/decode | |
| conditionings | CONDITIONING_LIST | List of conditionings from Smart Tile Conditioning | |
| negative | CONDITIONING | Negative conditioning | |
| seed | INT | 00โ18446744073709550000 | Random seed for sampling |
| steps | INT | 201โ100 | Number of sampling steps |
| cfg | FLOAT | 7.01โ30 | CFG scale |
| sampler_name | COMBO | euler | Sampler algorithm |
| scheduler | COMBO | normal | Noise scheduler |
| denoise | FLOAT | 0.400โ1 | Denoise strength |
| bundleopt | SMART_TILE_BUNDLE | Bundle from Smart Tile Calculator |
Outputs (2)
| Name | Type | Description |
|---|---|---|
| image | IMAGE | โ |
| segs | SEGS | โ |