Fix QwenImage _forward argument alignment for ComfyUI master（修复 QwenImage 在 ComfyUI master 下的参数错位问题）

[VipCrazyMaster]

Background / 背景

After upgrading to ComfyUI master, QwenImage-based nodes could crash
(e.g. 'list' object has no attribute 'dtype') or silently drift in generation
due to positional argument mismatch in _forward.

在升级到 ComfyUI master 之后，QwenImage 相关节点可能出现
'list' object has no attribute 'dtype' 报错，
或者由于 _forward 位置参数错位而导致生成语义漂移（模型不听 prompt）。

---

What this PR does / 本次修改内容

• Aligns _forward parameter order with ComfyUI master implementation
  对齐 _forward 的参数顺序，使其与 ComfyUI master 实现保持一致
• Removes legacy guidance handling from time_text_embed
  移除历史遗留的 guidance 注入逻辑，避免参数错位
• Safely unwraps single-element additional_t_cond without changing semantics
  仅对单元素 additional_t_cond 做安全解包，拒绝多元素以避免语义变化
• Adds bilingual (CN/EN) comments explaining the compatibility constraints
  增加中英文注释，说明与 ComfyUI master 兼容所需的约束条件

---

Why this is safe / 修改安全性说明

• Matches upstream comfy/ldm/qwen_image/model.py signatures exactly
  与上游 comfy/ldm/qwen_image/model.py 的函数签名完全一致
• Refuses to merge multi-element lists to avoid CFG / conditioning drift
  明确拒绝合并多元素列表，避免 CFG / 条件语义漂移
• No behavior change when inputs are already tensors
  当输入已是 tensor 时，不会引入任何行为变化

---

Tested / 测试环境

• ComfyUI master
• Python 3.12 / PyTorch 2.7
• QwenImage + ComfyUI-nunchaku

[VipCrazyMaster]

EN

Refuses to merge multi-element list/tuple for tensor-like inputs (e.g. additional_t_cond) to avoid CFG/conditioning semantic drift; multi-image inputs are handled via ref_latents as intended.

CN

仅对 additional_t_cond 等“tensor-like 输入”拒绝合并多元素 list/tuple，以避免 CFG/conditioning 语义漂移；多图输入按设计通过 ref_latents 传入并由模型内部逻辑处理。

[HelloClyde]

great! it works

[Michoko92]

Does your fix work also with Qwen edit models? For some reason, I still get the error with Qwen Image Edit 2509 if I put your file in the ComfyUI backend of SwarmUI.

[VipCrazyMaster]

Does your fix work also with Qwen edit models? For some reason, I still get the error with Qwen Image Edit 2509 if I put your file in the ComfyUI backend of SwarmUI.

SwarmUI没有试过，考虑是否是SwarmUI接口问题？因为我修复后在openwebui中是可以正常调用comfyui的。

[Michoko92]

Does your fix work also with Qwen edit models? For some reason, I still get the error with Qwen Image Edit 2509 if I put your file in the ComfyUI backend of SwarmUI.

SwarmUI没有试过，考虑是否是SwarmUI接口问题？因为我修复后在openwebui中是可以正常调用comfyui的。

I found the issue. It seems that SwarmUI keeps another copy of the Nunchaku nodes here:

SwarmUI\src\BuiltinExtensions\ComfyUIBackend\DLNodes\ComfyUI-nunchaku

I replaced the qwenimage.py in the models directory there too, and it works now! Thank you for the fix! 🙏

[DZFarouk7]

guys for the non savvy ,just run this commnd in the in omfyUI\custom_nodes\nunchaku_nodes git fetch origin pull/728/head:pr-728
git checkout pr-728

[istherea]

Thanks for sharing this! I was struggling with the exact same AttributeError after the recent ComfyUI update, and applying the PR #728 fix solved it perfectly for my QwenImageEdit-2509 workflow. Appreciate the help

[LV-Soft]

work!