Mikey Sampler Tiled Base Only
The auto-settings tiled hires pass
- base_model
- samples
- positive_cond_base
- negative_cond_base
- vae
- image
This is the plain-English version of Mikey Sampler Tiled Advanced Base Only - same core trick, upscale the image, split it into tiles, resample each one for extra detail, stitch it back together, decode straight to a finished image - but with the sampler internals hidden. No steps, no CFG, no sampler/scheduler dropdowns, no tile size. You get two dials: how much bigger, and how hard the tiled pass reworks what it finds.
How it works
Once your base sampling produces a latent, this node upscales it, tiles the result, resamples each tile to add detail the initial pass at lower resolution couldn't have generated, and blends the tiles back into one image - decoding for you along the way, since tiling has to happen in pixel space. It's the README's own framing for both tiled samplers in this pack: "these samplers use a tiled approach to resample the image after the upscale model has upscaled the image, to help enhance the image with more details." The Advanced sibling exposes every knob in that process; this one picks sensible values internally and just asks you for size and strength.
The inputs and outputs that matter
base_model,samples,positive_cond_base,negative_cond_base,vae- the base essentials, same as the rest of the Base Only family.model_name- the neural upscale model for the pre-tiling upscale step, pulled from yourupscale_modelsfolder - a real ESRGAN-style checkpoint, not a resize algorithm.seed- drives the sampling.upscale_by(default 1, min 0.1) - target size multiplier before the image gets tiled.tiler_denoise(default 0.25) - how much each tile's content gets reworked. This is the one strength dial you have; higher values add more detail per tile but raise the risk of visible seams where tiles disagree with their neighbors, which is exactly why the default sits on the conservative side.
Output is image - a finished, decoded IMAGE, not a latent. Like its Advanced sibling, tiling forces the decode to happen inside the node.
How to install it
Bundled with the rest of Mikey Nodes. ComfyUI Manager: search "Mikey Nodes," install, restart. Manually:
cd ComfyUI/custom_nodes
git clone https://github.com/bash-j/mikey_nodes
then restart. As with every sampler in this pack that has a model_name widget, you need an actual upscale model file in ComfyUI/models/upscale_models - get one from OpenModelDB. The author's own recommendation for a balanced, not-oversharpened result is lollipop.
Common issues & troubleshooting
Seams between tiles. Lower tiler_denoise first - it's the only strength control this node exposes, so it's also the only place to pull back from if tiles start disagreeing with each other visibly.
Wanting more control than this node offers. If you find yourself wishing you could touch steps, CFG, sampler, or tile size directly, that's the signal to move up to Mikey Sampler Tiled Advanced Base Only rather than fighting this node's two-dial design - it's deliberately the simpler option, not a cut-down one you're meant to work around.
Empty model_name dropdown. Reads from ComfyUI/models/upscale_models; an empty folder means nothing to pick from.
Expected a LATENT back. Same as its Advanced sibling - this decodes internally. If your graph needs to stay in latent space for another sampling stage, this isn't the node for that spot; a tiled pass is inherently a pixel-space operation, so both tiled variants in this pack hand back images, not latents.
Inputs (9)
| Name | Type | Default | Description |
|---|---|---|---|
| base_model | MODEL | — | |
| samples | LATENT | — | |
| positive_cond_base | CONDITIONING | — | |
| negative_cond_base | CONDITIONING | — | |
| vae | VAE | — | |
| model_name | COMBO | 0 options: | |
| seed | INT | 00–18446744073709550000 | — |
| upscale_by | FLOAT | 1.00.1–10 | — |
| tiler_denoise | FLOAT | 0.250–1 | — |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| image | IMAGE | — |