ZML_预设分辨率V3
Per-model resolution presets, so SDXL and Wan stop fighting over one dropdown
- 宽
- 高
- latent
V3 of ZML_预设分辨率 solves the problem you hit about three weeks after installing V1: your SDXL presets and your Wan presets and your Flux presets all live in one dropdown, and it's a mess. V3 adds a 模型预设 (model preset) selector so each model family gets its own resolution list, kept in its own folder. Flip to "SDXL," get the SDXL sizes; flip to "Wan," get the Wan sizes. No more scrolling past 1216x832 when you're looking for 1280x768.
The mechanism, straight from the source: the node scans zml_w/txt/Preset integer/ for subfolders, one per model preset. Each folder holds a resolutions.txt in the same W,H / Name_W,H format as the earlier versions. The "默认" (default) preset is always the shared Preset integer.txt, and you can define per-model random rules too. It's a filesystem-as-config setup, which is exactly the right call for this - adding a new model family is "make a folder, write a file."
The controls
- 模型预设 - enum: 默认 or 自定义 to start, plus any model folders you add.
- 分辨率预设 - the per-model resolution dropdown (same SDXL spread by default).
- 自定义宽 / 自定义高 - INT 64–16384, step 8 - available for the 自定义 branch.
- 互换宽高 / 随机模式 / 批次数量 - inherited from V1/V2.
Outputs: 宽, 高, latent (LATENT), same contract as its siblings.
Which version should you actually use?
Honestly: skip V1 and V2 and go straight here if you juggle more than one model family. V1 is fine if you're SDXL-only, V2 if you mostly want the custom fields, but V3 does both and adds the organization. It also matches how the author actually works - they built it because cramming every model's sizes into one list stopped scaling, and the README's changelog frames it as "so you don't have to squeeze SDXL and everything else together." The one thing to remember: presets live in files, so a new folder needs a restart (or a node reload) before it shows up in the dropdown.
Same shared caveats as every node in this pack: Chinese-first UI (grab the README's translation patch), single-maintainer project (issues/PRs on GitHub are the support path), and no model files or extra dependencies to install.
Install
ComfyUI Manager → search ComfyUI-ZML-Image, or:
cd ComfyUI/custom_nodes
git clone https://github.com/zml-w/ComfyUI-ZML-Image
restart, done.
Inputs (7)
| Name | Type | Default | Description |
|---|---|---|---|
| 模型预设 | COMBO | 默认 | 2 options: 默认, 自定义 |
| 分辨率预设 | COMBO | SDXL_768x1344 | 23 options: SDXL_768x1344, SDXL_768x1280, SDXL_832x1216, SDXL_832x1152, SDXL_896x1152, SDXL_896x1008, +17 |
| 自定义宽opt | INT | 102464–16384 | — |
| 自定义高opt | INT | 102464–16384 | — |
| 互换宽高opt | BOOLEAN | false | — |
| 随机模式opt | BOOLEAN | false | — |
| 批次数量opt | INT | 11–100 | — |
Outputs (3)
| Name | Type | Description |
|---|---|---|
| 宽 | INT | — |
| 高 | INT | — |
| latent | LATENT | — |