[Bug] qwen image 2509 error

[vvhitevvizard]

Checklist

• 1. I have searched for related issues and FAQs (https://github.com/nunchaku-tech/nunchaku/blob/main/docs/faq.md) but was unable to find a solution.
• 2. The issue persists in the latest version.
• 3. Please note that without environment information and a minimal reproducible example, it will be difficult for us to reproduce and address the issue, which may delay our response.
• 4. If your report is a question rather than a bug, please submit it as a discussion at https://github.com/nunchaku-tech/ComfyUI-nunchaku/discussions/new/choose. Otherwise, this issue will be closed.
• 5. I will do my best to describe the issue in English.

Describe the Bug

I use comfyui + nunchaku fp4 qwen image edit 2509 model. It was working before but recently it stopped.
I get this error:

Using pytorch attention in VAE VAE load device: cuda:0, offload device: cpu, dtype: torch.bfloat16 Requested to load WanVAE loaded completely; 11349.59 MB usable, 242.03 MB loaded, full load: True Found quantization metadata version 1 Using MixedPrecisionOps for text encoder CLIP/text encoder model load device: cuda:0, offload device: cpu, current: cpu, dtype: torch.float16 Requested to load QwenImageTEModel_ loaded completely; 13258.65 MB usable, 7910.29 MB loaded, full load: True Enabling CPU offload WARNING, you accessed scaled_fp8 from the model config object which doesn't exist. Please fix your code. model weight dtype torch.bfloat16, manual cast: None model_type FLUX Requested to load NunchakuQwenImage 0%| | 0/4 [00:00<?, ?it/s] !!! Exception during processing !!! 'list' object has no attribute 'dtype' Traceback (most recent call last): File "D:_ai\ComfyUI\execution.py", line 516, in execute output_data, output_ui, has_subgraph, has_pending_tasks = await get_output_data(prompt_id, unique_id, obj, input_data_all, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "D:_ai\ComfyUI\execution.py", line 330, in get_output_data return_values = await _async_map_node_over_list(prompt_id, unique_id, obj, input_data_all, obj.FUNCTION, allow_interrupt=True, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "D:_ai\ComfyUI\custom_nodes\comfyui-lora-manager\py\metadata_collector\metadata_hook.py", line 165, in async_map_node_over_list_with_metadata results = await original_map_node_over_list( ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ ...<2 lines>... ) ^ File "D:_ai\ComfyUI\execution.py", line 304, in _async_map_node_over_list await process_inputs(input_dict, i) File "D:_ai\ComfyUI\execution.py", line 292, in process_inputs result = f(**inputs) File "D:_ai\ComfyUI\nodes.py", line 1538, in sample return common_ksampler(model, seed, steps, cfg, sampler_name, scheduler, positive, negative, latent_image, denoise=denoise) File "D:_ai\ComfyUI\nodes.py", line 1505, in common_ksampler samples = comfy.sample.sample(model, noise, steps, cfg, sampler_name, scheduler, positive, negative, latent_image, denoise=denoise, disable_noise=disable_noise, start_step=start_step, last_step=last_step, force_full_denoise=force_full_denoise, noise_mask=noise_mask, callback=callback, disable_pbar=disable_pbar, seed=seed) File "D:_ai\ComfyUI\comfy\sample.py", line 60, in sample samples = sampler.sample(noise, positive, negative, cfg=cfg, latent_image=latent_image, start_step=start_step, last_step=last_step, force_full_denoise=force_full_denoise, denoise_mask=noise_mask, sigmas=sigmas, callback=callback, disable_pbar=disable_pbar, seed=seed) File "D:_ai\ComfyUI\comfy\samplers.py", line 1178, in sample return sample(self.model, noise, positive, negative, cfg, self.device, sampler, sigmas, self.model_options, latent_image=latent_image, denoise_mask=denoise_mask, callback=callback, disable_pbar=disable_pbar, seed=seed) File "D:_ai\ComfyUI\comfy\samplers.py", line 1068, in sample return cfg_guider.sample(noise, latent_image, sampler, sigmas, denoise_mask, callback, disable_pbar, seed) ~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "D:_ai\ComfyUI\comfy\samplers.py", line 1050, in sample output = executor.execute(noise, latent_image, sampler, sigmas, denoise_mask, callback, disable_pbar, seed, latent_shapes=latent_shapes) File "D:_ai\ComfyUI\comfy\patcher_extension.py", line 112, in execute return self.original(*args, **kwargs) ~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^ File "D:_ai\ComfyUI\comfy\samplers.py", line 994, in outer_sample output = self.inner_sample(noise, latent_image, device, sampler, sigmas, denoise_mask, callback, disable_pbar, seed, latent_shapes=latent_shapes) File "D:_ai\ComfyUI\comfy\samplers.py", line 980, in inner_sample samples = executor.execute(self, sigmas, extra_args, callback, noise, latent_image, denoise_mask, disable_pbar) File "D:_ai\ComfyUI\comfy\patcher_extension.py", line 112, in execute return self.original(*args, **kwargs) ~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^ File "D:_ai\ComfyUI\comfy\samplers.py", line 752, in sample samples = self.sampler_function(model_k, noise, sigmas, extra_args=extra_args, callback=k_callback, disable=disable_pbar, **self.extra_options) File "D:_ai\python_embeded\Lib\torch_cu\torch\utils_contextlib.py", line 120, in decorate_context return func(*args, **kwargs) File "D:_ai\ComfyUI\comfy\k_diffusion\sampling.py", line 199, in sample_euler denoised = model(x, sigma_hat * s_in, **extra_args) File "D:_ai\ComfyUI\comfy\samplers.py", line 401, in call out = self.inner_model(x, sigma, model_options=model_options, seed=seed) File "D:_ai\ComfyUI\comfy\samplers.py", line 953, in call return self.outer_predict_noise(*args, **kwargs) ~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^ File "D:_ai\ComfyUI\comfy\samplers.py", line 960, in outer_predict_noise ).execute(x, timestep, model_options, seed) ~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "D:_ai\ComfyUI\comfy\patcher_extension.py", line 112, in execute return self.original(*args, **kwargs) ~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^ File "D:_ai\ComfyUI\comfy\samplers.py", line 963, in predict_noise return sampling_function(self.inner_model, x, timestep, self.conds.get("negative", None), self.conds.get("positive", None), self.cfg, model_options=model_options, seed=seed) File "D:_ai\ComfyUI\comfy\samplers.py", line 381, in sampling_function out = calc_cond_batch(model, conds, x, timestep, model_options) File "D:_ai\ComfyUI\comfy\samplers.py", line 206, in calc_cond_batch return _calc_cond_batch_outer(model, conds, x_in, timestep, model_options) File "D:_ai\ComfyUI\comfy\samplers.py", line 214, in _calc_cond_batch_outer return executor.execute(model, conds, x_in, timestep, model_options) ~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "D:_ai\ComfyUI\comfy\patcher_extension.py", line 112, in execute return self.original(*args, **kwargs) ~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^ File "D:_ai\ComfyUI\comfy\samplers.py", line 326, in calc_cond_batch output = model.apply_model(input_x, timestep, **c).chunk(batch_chunks) ~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^ File "D:_ai\ComfyUI\comfy\model_base.py", line 162, in apply_model return comfy.patcher_extension.WrapperExecutor.new_class_executor( ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ...<2 lines>... comfy.patcher_extension.get_all_wrappers(comfy.patcher_extension.WrappersMP.APPLY_MODEL, transformer_options) ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ).execute(x, t, c_concat, c_crossattn, control, transformer_options, **kwargs) ~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "D:_ai\ComfyUI\comfy\patcher_extension.py", line 112, in execute return self.original(*args, **kwargs) ~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^ File "D:_ai\ComfyUI\comfy\model_base.py", line 204, in _apply_model model_output = self.diffusion_model(xc, t, context=context, control=control, transformer_options=transformer_options, **extra_conds) File "D:_ai\python_embeded\Lib\torch_cu\torch\nn\modules\module.py", line 1775, in _wrapped_call_impl return self._call_impl(*args, **kwargs) ~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^ File "D:_ai\python_embeded\Lib\torch_cu\torch\nn\modules\module.py", line 1786, in _call_impl return forward_call(*args, **kwargs) File "D:_ai\ComfyUI\comfy\ldm\qwen_image\model.py", line 411, in forward return comfy.patcher_extension.WrapperExecutor.new_class_executor( ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ...<2 lines>... comfy.patcher_extension.get_all_wrappers(comfy.patcher_extension.WrappersMP.DIFFUSION_MODEL, transformer_options) ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ).execute(x, timestep, context, attention_mask, ref_latents, additional_t_cond, transformer_options, **kwargs) ~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "D:_ai\ComfyUI\comfy\patcher_extension.py", line 112, in execute return self.original(*args, **kwargs) ~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^ File "D:_ai\ComfyUI\custom_nodes\ComfyUI-nunchaku\models\qwenimage.py", line 726, in _forward else self.time_text_embed(timestep, guidance, hidden_states) ~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "D:_ai\python_embeded\Lib\torch_cu\torch\nn\modules\module.py", line 1775, in _wrapped_call_impl return self._call_impl(*args, **kwargs) ~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^ File "D:_ai\python_embeded\Lib\torch_cu\torch\nn\modules\module.py", line 1786, in _call_impl return forward_call(*args, **kwargs) File "D:_ai\ComfyUI\comfy\ldm\qwen_image\model.py", line 81, in forward timesteps_emb = self.timestep_embedder(timesteps_proj.to(dtype=hidden_states.dtype)) ^^^^^^^^^^^^^^^^^^^ AttributeError: 'list' object has no attribute 'dtype' Prompt executed in 28.31 seconds

Environment

Windows 11, python 13, Torch 2.9, Cu130

Reproduction Steps

the error is reproduced using the default nuchaku's workflow

[vvhitevvizard]

gemini:
FIX: ComfyUI Core expects (timestep, hidden_states).
We must extract the tensor from the list and NOT pass guidance as an argument.

file ComfyUI-nunchaku\models\qwenimage.py:

    temb = (
        self.time_text_embed(timestep, hidden_states)
        if guidance is None
        else self.time_text_embed(timestep, guidance, hidden_states)
    )

->

ref_tensor = hidden_states[0] if isinstance(hidden_states, list) else hidden_states
temb = self.time_text_embed(timestep, ref_tensor)

[frost-byte]

gemini: FIX: ComfyUI Core expects (timestep, hidden_states). We must extract the tensor from the list and NOT pass guidance as an argument.

file ComfyUI-nunchaku\models\qwenimage.py:

    temb = (
        self.time_text_embed(timestep, hidden_states)
        if guidance is None
        else self.time_text_embed(timestep, guidance, hidden_states)
    )

->

ref_tensor = hidden_states[0] if isinstance(hidden_states, list) else hidden_states
temb = self.time_text_embed(timestep, ref_tensor)

I am just trying to following along with what you're saying, are you talking about the lines here?

ComfyUI-nunchaku/models/qwenimage.py

Line 723 in dbfd194

temb = (

[trollver9000]

Ive got this error without loading any lora - just by loading qwen image edit and trying to infer, this happened after i updated comfy

Its very annoying that everytime i udpate comfy , this nunchaku node is broken instantly. what is going on ?

~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

File "M:\ComfyUI_windows_portable\ComfyUI\comfy\model_management.py", line 509, in model_load
self.model_use_more_vram(use_more_vram, force_patch_weights=force_patch_weights)
~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "M:\ComfyUI_windows_portable\ComfyUI\comfy\model_management.py", line 539, in model_use_more_vram
return self.model.partially_load(self.device, extra_memory, force_patch_weights=force_patch_weights)
~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "M:\ComfyUI_windows_portable\ComfyUI\comfy\model_patcher.py", line 980, in partially_load
self.detach()
~~~~~~~~~~~^^
File "M:\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI-nunchaku\model_patcher.py", line 41, in detach
self.model.diffusion_model.to_safely(self.offload_device)
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^
File "M:\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI-QwenImageLoraLoader\wrappers\qwenimage.py", line 65, in to_safely
self.model.to(device)
^^^^^^^^^^^^^
AttributeError: 'NoneType' object has no attribute 'to'

[trollver9000]

DOWNGRADING TO COMFY 4.0 FROM 2 WEEKS AGO SOLVES THIS, SO THIS IS SOMETHING THAT COMFY CORE CODE IS RESPONSIBLE FOR BUT qwen LORAS DO NOT WORK ( no attribute pinned) m and this is an issue with this nunchaku dev branch

[thefirstangel3d]

i use my comfy setup in a venv configuration, so i activated my venv then reverted to 0.4.0, in console using

git checkout 791e30f

then restart comfy. so both nunchaku qwen and loras worked fine after.

for portable, go to ...\ComfyUI_windows_portable\comfyui in explorer adress list type cmd (+enter) and in console type

git fetch --all
git checkout v0.4.0

should work for comfyUI_windows portable
hope this helps.

[frost-byte]

git checkout 791e30f

then restart comfy. so both nunchaku qwen and loras worked fine after.

for portable, go to ...\ComfyUI_windows_portable\comfyui in explorer adress list type cmd (+enter) and in console type

git fetch --all git checkout v0.4.0

Did you change the front end version too?
The hard part for me is figuring out which version of the ui to switch back to.

[thefirstangel3d]

my current (functional) pip list gives me:

comfyui-embedded-docs 0.3.1
comfyui_frontend_package 1.34.9
comfyui-manager 4.0.3b7
comfyui_workflow_templates 0.7.60
comfyui-workflow-templates-core 0.3.46
comfyui-workflow-templates-media-api 0.3.23
comfyui-workflow-templates-media-image 0.3.38
comfyui-workflow-templates-media-other 0.3.50
comfyui-workflow-templates-media-video 0.3.19

[thefirstangel3d]

and yes, dependency hell isn't fun and this is a temporary fix, i'm sure the devs already work on the issue. seen plenty of nunchaku/qwen users run into this in recent update.