[GH-ISSUE #114] EmptyQwenImageLayeredLatentImage: 11 locales still describe `layers` wrong

Originally created by @christian-byrne on GitHub (Aug 6, 2026). Original GitHub issue: https://github.com/Comfy-Org/embedded-docs/issues/114 ## Summary `comfyui_embedded_docs/docs/EmptyQwenImageLayeredLatentImage/` has 12 locales. The English source has been corrected in https://github.com/Comfy-Org/embedded-docs/pull/113; the other 11 still carry a description of the `layers` widget that is actively misleading, and none of them state what the extra slot in the output shape is. ## What is wrong Every locale describes `layers` as some variant of "the number of additional layers to add to the latent structure, defining the depth of the latent representation", and every locale prints the output shape `[batch_size, 16, layers + 1, height // 8, width // 8]` without explaining the `+ 1`. The `+ 1` is not padding or depth. Qwen-Image-Layered regenerates the full image in addition to the layers, so slot 0 is the full image, it is not one of the layers, and it should be discarded when only the layers are wanted. comfyanonymous, verbatim: > it's a model specific thing, it's because this model is an image generation model that regenerates the full image and the layers > > so if you set layers to zero you get only the original image > > but if you set layers to 2 you get the original image + 2 layers > > the first image is exactly the same as the original and should be discarded > > and all the layers when composited together should recreate the original image Verified against source: `comfy_extras/nodes_qwen.py:109-130` allocates `torch.zeros([batch_size, 16, layers + 1, height // 8, width // 8])`, and `layers` has `min=0`. ## Files needing the same fix as `en.md` - [ ] `ar.md` - [ ] `es.md` - [ ] `fa.md` - [ ] `fr.md` - [ ] `ja.md` - [ ] `ko.md` - [ ] `pt-BR.md` - [ ] `ru.md` - [ ] `tr.md` - [ ] `zh.md` - [ ] `zh-TW.md` All under `comfyui_embedded_docs/docs/EmptyQwenImageLayeredLatentImage/`. ## Why this was not done in the same PR `.cursorrules` in this repo is explicit: > Do not automatically start translation work for other language files unless actively requested by the user Machine-translating 11 locales without a reviewer who reads them would be worse than leaving them stale and tracked. Hence this issue. ## Note for whoever picks this up `ja.md`, `ko.md` and `zh.md` for this node all carry source fingerprint `99497e3e…` while `en.md` carries `fe979666…`, so the translations were already generated from a different English revision than the one currently in the repo. That is not specific to this node: across the repo, 540 of 739 node directories that have both `en.md` and `zh.md` have mismatched fingerprints. Worth a separate look at whether the fingerprint is meant to be a staleness signal, because right now it is not being used as one.

bsakaguchi commented

2026-10-26 22:08:08 +00:00

Owner

Originally created by @christian-byrne on GitHub (Aug 6, 2026).
Original GitHub issue: https://github.com/Comfy-Org/embedded-docs/issues/114

Summary

comfyui_embedded_docs/docs/EmptyQwenImageLayeredLatentImage/ has 12 locales. The English source has been corrected in https://github.com/Comfy-Org/embedded-docs/pull/113; the other 11 still carry a description of the layers widget that is actively misleading, and none of them state what the extra slot in the output shape is.

What is wrong

Every locale describes layers as some variant of "the number of additional layers to add to the latent structure, defining the depth of the latent representation", and every locale prints the output shape [batch_size, 16, layers + 1, height // 8, width // 8] without explaining the + 1.

The + 1 is not padding or depth. Qwen-Image-Layered regenerates the full image in addition to the layers, so slot 0 is the full image, it is not one of the layers, and it should be discarded when only the layers are wanted. comfyanonymous, verbatim:

it's a model specific thing, it's because this model is an image generation model that regenerates the full image and the layers

so if you set layers to zero you get only the original image

but if you set layers to 2 you get the original image + 2 layers

the first image is exactly the same as the original and should be discarded

and all the layers when composited together should recreate the original image

Verified against source: comfy_extras/nodes_qwen.py:109-130 allocates torch.zeros([batch_size, 16, layers + 1, height // 8, width // 8]), and layers has min=0.

Files needing the same fix as `en.md`

ar.md
es.md
fa.md
fr.md
ja.md
ko.md
pt-BR.md
ru.md
tr.md
zh.md
zh-TW.md

All under comfyui_embedded_docs/docs/EmptyQwenImageLayeredLatentImage/.

Why this was not done in the same PR

.cursorrules in this repo is explicit:

Do not automatically start translation work for other language files unless actively requested by the user

Machine-translating 11 locales without a reviewer who reads them would be worse than leaving them stale and tracked. Hence this issue.

Note for whoever picks this up

ja.md, ko.md and zh.md for this node all carry source fingerprint 99497e3e… while en.md carries fe979666…, so the translations were already generated from a different English revision than the one currently in the repo. That is not specific to this node: across the repo, 540 of 739 node directories that have both en.md and zh.md have mismatched fingerprints. Worth a separate look at whether the fingerprint is meant to be a staleness signal, because right now it is not being used as one.