String Resolution Jimeng

Here's the annoying part about Jimeng resolution labels: they're not a property of the model, they're a property of the node you call. This selector outputs the union of them, which means it can hand you a tier the next node in the chain doesn't accept. Read the second half of this article before you wire it up.

What it is, and when you'd want it

String Resolution Jimeng matches an image to one of three labels - 1k, 2k, 4k - and outputs it as a string. No resizing, no generation, no API key. It's the label half of a two-node split; its pixel-half partner in this pack is Image Resize Jimeng.

Jimeng is ByteDance's consumer image app (the international version is Dreamina), and the KB's read on ByteDance applies exactly: everything flagship stays on the API, and what ships openly ships as components under their own names. Jimeng definitely lives on the API side. If you run the family's ComfyUI nodes - they live in a different pack, ComfyUI-1hewJimeng - this node produces the resolution string those nodes want.

Worth knowing if you're deciding whether to bother: Jimeng barely registers in the English-language subreddits. A corpus count for the term since the start of 2025 lands at seventeen threads, most of them passing mentions from Chinese-market users - versus the constant churn around Seedream and Nano Banana. The tooling exists, it's just not something Reddit will have opinions about for you.

How the matching works

Area, not shape: width * height against each tier's target area, compared in log space, nearest wins. Targets are each tier's official 1:1 size, and the 1k one is unusual:

| Tier | 1:1 official size | Target area | Upper edge | | --- | --- | --- | --- | | 1k | 1328x1328 | 1,763,584 | ~2.72 MP | | 2k | 2048x2048 | 4,194,304 | ~8.39 MP | | 4k | 4096x4096 | 16,777,216 | - |

Which gives some clean examples:

• 1920x1080 = 2.07 MP, comfortably under the 2.72 MP edge → 1k. Jimeng's 1k is a fatter tier than everyone else's 1k (1328², not 1024²), so more images fall into it.
• 2048x2048 → 2k.
• 3840x2160 = 8.29 MP, just under the 8.39 MP edge → 2k. The 4K-labelled file does not get the 4k tier.
• 4096x4096 → 4k.

Batches are matched per frame, one label per line.

Inputs and output

selection is the COMBO, default 1k - the manual fallback. Leave image unconnected and it's a string constant, which is what you want for text-to-image.

image is the optional IMAGE input; connect it and the tier is inferred per frame.

string is the output, a STRING for the Jimeng node's resolution input.

The mismatch nobody warns you about

Because this node emits the union of Jimeng's tiers, it can produce a label the node you're feeding doesn't support. From the author's docs, the availability is all over the place:

• Jimeng Image T2I takes 1k / 2k / 4k.
• Jimeng Image Multi Edit takes 2k / 4k only.
• The 4.x line (4.0 through 4.7) and 5.0 Lite take 2k / 4k.
• Jimeng Image 3.0, its ControlNet and IP variants, and the edit/style/subject nodes take 1k / 2k.

So a 1080p reference feeding a 4.x node produces 1k and gets rejected. Two honest fixes: set selection by hand for that chain, or gate the tier with a logic node before it reaches the API node. Also note the 2.0_pro model is out of scope entirely - it isn't tiered by 1k/2k/4k at all, it uses its own 1024-base size table, which is exactly why Image Resize Jimeng's auto mode excludes those presets too.

Installing it

ComfyUI Manager → 1hewNodes → install → restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/1hew/ComfyUI-1hewNodes

requirements.txt installs a full CV stack - opencv-python, scikit-image, psd-tools, color-matcher, ultralytics, rembg, transparent-background, onnxruntime, av, torchaudio - for a pack that also handles masks, video, detection and tiling. Registration runs through the newer comfy_api.latest schema API, so keep ComfyUI current, and note that a failed module import is logged and skipped: the symptom is a node missing from the menu, not a crash.

Where people get burned

This is the label, not the pixels. Use Image Resize Jimeng for actual dimensions, and remember auto there deliberately skips 2.0_pro.

Tier is the cost knob on an API call, so a 4k label on a 2 MP source is money spent on nothing.

Check the union problem above before blaming the node for an API error.

Old workflows may show a missing node - v3.28.0 swapped the generic resolution selector for per-model ones, this being one of them.