Repository navigation
Please add gguf‑quantized support for "Krea‑2" #464
Description
Activity
- changed the title
[-]Please add gguf‑quantized support for Krea‑2[/-][+]Please add gguf‑quantized support for "Krea‑2"[/+]on Jun 24, 2026 My gguf nodes are a forked version of these for krea-2, if anyone knows how to merge my loaders py into the normal gguf repo then we could all already have it. I coded it in last night. Its on my repo but i dont know how to use github at all to merge it myself. This site is like a foreign language for some reason to me
#463 (comment)
https://github.com/m8rr/ComfyUI-GGUF/tree/Dynamic-VRAM@RealRebelAI Your GGUF works fine in my environment, though I'm not sure if the calculations are correct. The GGUF from the link provided by @makisekurisu-jp throws a [WARNING] unet unexpected: ['last.down.weight', 'last.up.weight'], but it does work.
And ComfyUI's official BF16 doesn't have ['last.down.weight', 'last.up.weight'], so it seems safe to ignore.
ideogram4, boogu, krea2 / UNet and TE both function properly using GGUF.
While Gemma-4 E2B appears to have some problems, it operates well in my specific environment, including vision recognition.#463 (comment)
https://github.com/m8rr/ComfyUI-GGUF/tree/Dynamic-VRAM@RealRebelAI Your GGUF works fine in my environment, though I'm not sure if the calculations are correct. The GGUF from the link provided by @makisekurisu-jp throws a [WARNING] unet unexpected: ['last.down.weight', 'last.up.weight'], but it does work.
And ComfyUI's official BF16 doesn't have ['last.down.weight', 'last.up.weight'], so it seems safe to ignore.
ideogram4, boogu, krea2 / UNet and TE both function properly using GGUF.
While Gemma-4 E2B appears to have some problems, it operates well in my specific environment, including vision recognition.My gguf nodes are a forked version of these for krea-2, if anyone knows how to merge my loaders py into the normal gguf repo then we could all already have it. I coded it in last night. Its on my repo but i dont know how to use github at all to merge it myself. This site is like a foreign language for some reason to me
Do you also get the following warning when running your quantized GGUF model?
[WARNING] unet unexpected: ['last.down.weight', 'last.up.weight']
I encountered the same warning when using a model quantized with vantagewithai.
You can only open a PR; merging into the main branch requires access to the city96 repository. You’ll need to fork his repository and maintain your own fork.My gguf nodes are a forked version of these for krea-2, if anyone knows how to merge my loaders py into the normal gguf repo then we could all already have it. I coded it in last night. Its on my repo but i dont know how to use github at all to merge it myself. This site is like a foreign language for some reason to me
Do you also get the following warning when running your quantized GGUF model? [WARNING] unet unexpected: ['last.down.weight', 'last.up.weight'] I encountered the same warning when using a model quantized with vantagewithai. You can only open a PR; merging into the main branch requires access to the city96 repository. You’ll need to fork his repository and maintain your own fork.
The last.down.weight issue warning is just kreas layering format forcing a layer that comfy doesnt accept so it bypasses them essentially
My gguf nodes are a forked version of these for krea-2, if anyone knows how to merge my loaders py into the normal gguf repo then we could all already have it. I coded it in last night. Its on my repo but i dont know how to use github at all to merge it myself. This site is like a foreign language for some reason to me
Do you also get the following warning when running your quantized GGUF model? [WARNING] unet unexpected: ['last.down.weight', 'last.up.weight'] I encountered the same warning when using a model quantized with vantagewithai. You can only open a PR; merging into the main branch requires access to the city96 repository. You’ll need to fork his repository and maintain your own fork.
The last.down.weight issue warning is just kreas layering format forcing a layer that comfy doesnt accept so it bypasses them essentially
any idea why i go clip warning ? are wrong quant or model for clip ? or my comfyui need update cause already update and even disable city96 and calcius node.
[INFO] gguf qtypes: F32 (167), F16 (1), Q5_K (63), Q2_K (201) [INFO] model weight dtype torch.bfloat16, manual cast: torch.float32 [INFO] model_type FLUX [WARNING] unet unexpected: ['last.down.weight', 'last.up.weight'] [INFO] Requested to load Krea2 FETCH ComfyRegistry Data: 145/156 FETCH ComfyRegistry Data: 150/156 FETCH ComfyRegistry Data: 155/156 FETCH ComfyRegistry Data [DONE] [INFO] [ComfyUI-Manager] default cache updated: https://api.comfy.org/nodes FETCH DATA from: https://raw.githubusercontent.com/ltdrdata/ComfyUI-Manager/main/custom-node-list.json [DONE] [INFO] [ComfyUI-Manager] All startup tasks have been completed. [INFO] loaded partially; 1976.12 MB usable, 1755.61 MB loaded, 3284.59 MB offloaded, 220.50 MB buffer reserved, lowvram patches: 0 0%| | 0/4 [00:00<?, ?it/s, Model Initializing ... ] [ERROR] !!! Exception during processing !!! Krea2 expects conditioning with 12x2560=30720 features (a 12-layer Qwen3-VL stack) but got 2560. Load the text encoder with CLIPLoader type 'krea2'. [ERROR] Traceback (most recent call last): File "C:\comfui\ComfyUI_windows_portable\ComfyUI\execution.py", line 542, 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 "C:\comfui\ComfyUI_windows_portable\ComfyUI\execution.py", line 341, 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 "C:\comfui\ComfyUI_windows_portable\ComfyUI\execution.py", line 315, in _async_map_node_over_list await process_inputs(input_dict, i) File "C:\comfui\ComfyUI_windows_portable\ComfyUI\execution.py", line 303, in process_inputs result = f(**inputs) ^^^^^^^^^^^ File "C:\comfui\ComfyUI_windows_portable\ComfyUI\nodes.py", line 1584, in sample return common_ksampler(model, seed, steps, cfg, sampler_name, scheduler, positive, negative, latent_image, denoise=denoise) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\comfui\ComfyUI_windows_portable\ComfyUI\nodes.py", line 1548, in common_ksampler samples = comfy.sample.sample(model, noise, steps, cfg, sampler_name, scheduler, positive, negative, latent_image, ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\comfui\ComfyUI_windows_portable\ComfyUI\comfy\sample.py", line 74, 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 "C:\comfui\ComfyUI_windows_portable\ComfyUI\comfy\samplers.py", line 1444, 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 "C:\comfui\ComfyUI_windows_portable\ComfyUI\comfy\samplers.py", line 1334, in sample return cfg_guider.sample(noise, latent_image, sampler, sigmas, denoise_mask, callback, disable_pbar, seed) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\comfui\ComfyUI_windows_portable\ComfyUI\comfy\samplers.py", line 1316, in sample output = executor.execute(noise, latent_image, sampler, sigmas, denoise_mask, callback, disable_pbar, seed, latent_shapes=latent_shapes) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\comfui\ComfyUI_windows_portable\ComfyUI\comfy\patcher_extension.py", line 113, in execute return self.original(*args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\comfui\ComfyUI_windows_portable\ComfyUI\comfy\samplers.py", line 1254, in outer_sample output = self.inner_sample(noise, latent_image, device, sampler, sigmas, denoise_mask, callback, disable_pbar, seed, latent_shapes=latent_shapes) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\comfui\ComfyUI_windows_portable\ComfyUI\comfy\samplers.py", line 1229, in inner_sample samples = executor.execute(self, sigmas, extra_args, callback, noise, latent_image, denoise_mask, disable_pbar) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\comfui\ComfyUI_windows_portable\ComfyUI\comfy\patcher_extension.py", line 113, in execute return self.original(*args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\comfui\ComfyUI_windows_portable\ComfyUI\comfy\samplers.py", line 999, in sample samples = self.sampler_function(model_k, noise, sigmas, extra_args=extra_args, callback=k_callback, disable=disable_pbar, **self.extra_options) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\comfui\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\utils\_contextlib.py", line 124, in decorate_context return func(*args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^ File "C:\comfui\ComfyUI_windows_portable\ComfyUI\comfy\k_diffusion\sampling.py", line 205, in sample_euler denoised = model(x, sigma_hat * s_in, **extra_args) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\comfui\ComfyUI_windows_portable\ComfyUI\comfy\samplers.py", line 639, in __call__ out = self.inner_model(x, sigma, model_options=model_options, seed=seed) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\comfui\ComfyUI_windows_portable\ComfyUI\comfy\samplers.py", line 1202, in __call__ return self.outer_predict_noise(*args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\comfui\ComfyUI_windows_portable\ComfyUI\comfy\samplers.py", line 1209, in outer_predict_noise ).execute(x, timestep, model_options, seed) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\comfui\ComfyUI_windows_portable\ComfyUI\comfy\patcher_extension.py", line 113, in execute return self.original(*args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\comfui\ComfyUI_windows_portable\ComfyUI\comfy\samplers.py", line 1212, 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 "C:\comfui\ComfyUI_windows_portable\ComfyUI\comfy\samplers.py", line 619, in sampling_function out = calc_cond_batch(model, conds, x, timestep, model_options) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\comfui\ComfyUI_windows_portable\ComfyUI\comfy\samplers.py", line 210, in calc_cond_batch return _calc_cond_batch_outer(model, conds, x_in, timestep, model_options) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\comfui\ComfyUI_windows_portable\ComfyUI\comfy\samplers.py", line 218, in _calc_cond_batch_outer return executor.execute(model, conds, x_in, timestep, model_options) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\comfui\ComfyUI_windows_portable\ComfyUI\comfy\patcher_extension.py", line 113, in execute return self.original(*args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\comfui\ComfyUI_windows_portable\ComfyUI\comfy\samplers.py", line 334, in _calc_cond_batch output = model.apply_model(input_x, timestep_, **c).chunk(batch_chunks) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\comfui\ComfyUI_windows_portable\ComfyUI\comfy\model_base.py", line 191, in apply_model return comfy.patcher_extension.WrapperExecutor.new_class_executor( ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\comfui\ComfyUI_windows_portable\ComfyUI\comfy\patcher_extension.py", line 113, in execute return self.original(*args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\comfui\ComfyUI_windows_portable\ComfyUI\comfy\model_base.py", line 235, in _apply_model model_output = self.diffusion_model(xc, t, context=context, control=control, transformer_options=transformer_options, **extra_conds) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\comfui\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1778, in _wrapped_call_impl return self._call_impl(*args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\comfui\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1789, in _call_impl return forward_call(*args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\comfui\ComfyUI_windows_portable\ComfyUI\comfy\ldm\krea2\model.py", line 225, in forward return comfy.patcher_extension.WrapperExecutor.new_class_executor( ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\comfui\ComfyUI_windows_portable\ComfyUI\comfy\patcher_extension.py", line 113, in execute return self.original(*args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\comfui\ComfyUI_windows_portable\ComfyUI\comfy\ldm\krea2\model.py", line 244, in _forward context = self._unpack_context(context) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\comfui\ComfyUI_windows_portable\ComfyUI\comfy\ldm\krea2\model.py", line 285, in _unpack_context raise ValueError( ValueError: Krea2 expects conditioning with 12x2560=30720 features (a 12-layer Qwen3-VL stack) but got 2560. Load the text encoder with CLIPLoader type 'krea2'.[INFO] Prompt executed in 131.78 seconds
My gguf nodes are a forked version of these for krea-2, if anyone knows how to merge my loaders py into the normal gguf repo then we could all already have it. I coded it in last night. Its on my repo but i dont know how to use github at all to merge it myself. This site is like a foreign language for some reason to me
Do you also get the following warning when running your quantized GGUF model? [WARNING] unet unexpected: ['last.down.weight', 'last.up.weight'] I encountered the same warning when using a model quantized with vantagewithai. You can only open a PR; merging into the main branch requires access to the city96 repository. You’ll need to fork his repository and maintain your own fork.
The last.down.weight issue warning is just kreas layering format forcing a layer that comfy doesnt accept so it bypasses them essentially
any idea why i go clip warning ? are wrong quant or model for clip ? or my comfyui need update cause already update and even disable city96 and calcius node.
```
[INFO] gguf qtypes: F32 (167), F16 (1), Q5_K (63), Q2_K (201)
[INFO] model weight dtype torch.bfloat16, manual cast: torch.float32
[INFO] model_type FLUX
[WARNING] unet unexpected: ['last.down.weight', 'last.up.weight']
[INFO] Requested to load Krea2
FETCH ComfyRegistry Data: 145/156
FETCH ComfyRegistry Data: 150/156
FETCH ComfyRegistry Data: 155/156
FETCH ComfyRegistry Data [DONE]
[INFO] [ComfyUI-Manager] default cache updated: https://api.comfy.org/nodes
FETCH DATA from: https://raw.githubusercontent.com/ltdrdata/ComfyUI-Manager/main/custom-node-list.json [DONE]
[INFO] [ComfyUI-Manager] All startup tasks have been completed.
[INFO] loaded partially; 1976.12 MB usable, 1755.61 MB loaded, 3284.59 MB offloaded, 220.50 MB buffer reserved, lowvram patches: 0
0%| | 0/4 [00:00<?, ?it/s, Model Initializing ... ]
[ERROR] !!! Exception during processing !!! Krea2 expects conditioning with 12x2560=30720 features (a 12-layer Qwen3-VL stack) but got 2560. Load the text encoder with CLIPLoader type 'krea2'.
[ERROR] Traceback (most recent call last):
File "C:\comfui\ComfyUI_windows_portable\ComfyUI\execution.py", line 542, 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 "C:\comfui\ComfyUI_windows_portable\ComfyUI\execution.py", line 341, 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 "C:\comfui\ComfyUI_windows_portable\ComfyUI\execution.py", line 315, in _async_map_node_over_list
await process_inputs(input_dict, i)
File "C:\comfui\ComfyUI_windows_portable\ComfyUI\execution.py", line 303, in process_inputs
result = f(**inputs)
^^^^^^^^^^^
File "C:\comfui\ComfyUI_windows_portable\ComfyUI\nodes.py", line 1584, in sample
return common_ksampler(model, seed, steps, cfg, sampler_name, scheduler, positive, negative, latent_image, denoise=denoise)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\comfui\ComfyUI_windows_portable\ComfyUI\nodes.py", line 1548, in common_ksampler
samples = comfy.sample.sample(model, noise, steps, cfg, sampler_name, scheduler, positive, negative, latent_image,
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\comfui\ComfyUI_windows_portable\ComfyUI\comfy\sample.py", line 74, 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 "C:\comfui\ComfyUI_windows_portable\ComfyUI\comfy\samplers.py", line 1444, 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 "C:\comfui\ComfyUI_windows_portable\ComfyUI\comfy\samplers.py", line 1334, in sample
return cfg_guider.sample(noise, latent_image, sampler, sigmas, denoise_mask, callback, disable_pbar, seed)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\comfui\ComfyUI_windows_portable\ComfyUI\comfy\samplers.py", line 1316, in sample
output = executor.execute(noise, latent_image, sampler, sigmas, denoise_mask, callback, disable_pbar, seed, latent_shapes=latent_shapes)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\comfui\ComfyUI_windows_portable\ComfyUI\comfy\patcher_extension.py", line 113, in execute
return self.original(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\comfui\ComfyUI_windows_portable\ComfyUI\comfy\samplers.py", line 1254, in outer_sample
output = self.inner_sample(noise, latent_image, device, sampler, sigmas, denoise_mask, callback, disable_pbar, seed, latent_shapes=latent_shapes)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\comfui\ComfyUI_windows_portable\ComfyUI\comfy\samplers.py", line 1229, in inner_sample
samples = executor.execute(self, sigmas, extra_args, callback, noise, latent_image, denoise_mask, disable_pbar)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\comfui\ComfyUI_windows_portable\ComfyUI\comfy\patcher_extension.py", line 113, in execute
return self.original(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\comfui\ComfyUI_windows_portable\ComfyUI\comfy\samplers.py", line 999, in sample
samples = self.sampler_function(model_k, noise, sigmas, extra_args=extra_args, callback=k_callback, disable=disable_pbar, **self.extra_options)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\comfui\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\utils_contextlib.py", line 124, in decorate_context
return func(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^
File "C:\comfui\ComfyUI_windows_portable\ComfyUI\comfy\k_diffusion\sampling.py", line 205, in sample_euler
denoised = model(x, sigma_hat * s_in, **extra_args)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\comfui\ComfyUI_windows_portable\ComfyUI\comfy\samplers.py", line 639, in call
out = self.inner_model(x, sigma, model_options=model_options, seed=seed)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\comfui\ComfyUI_windows_portable\ComfyUI\comfy\samplers.py", line 1202, in call
return self.outer_predict_noise(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\comfui\ComfyUI_windows_portable\ComfyUI\comfy\samplers.py", line 1209, in outer_predict_noise
).execute(x, timestep, model_options, seed)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\comfui\ComfyUI_windows_portable\ComfyUI\comfy\patcher_extension.py", line 113, in execute
return self.original(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\comfui\ComfyUI_windows_portable\ComfyUI\comfy\samplers.py", line 1212, 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 "C:\comfui\ComfyUI_windows_portable\ComfyUI\comfy\samplers.py", line 619, in sampling_function
out = calc_cond_batch(model, conds, x, timestep, model_options)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\comfui\ComfyUI_windows_portable\ComfyUI\comfy\samplers.py", line 210, in calc_cond_batch
return _calc_cond_batch_outer(model, conds, x_in, timestep, model_options)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\comfui\ComfyUI_windows_portable\ComfyUI\comfy\samplers.py", line 218, in _calc_cond_batch_outer
return executor.execute(model, conds, x_in, timestep, model_options)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\comfui\ComfyUI_windows_portable\ComfyUI\comfy\patcher_extension.py", line 113, in execute
return self.original(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\comfui\ComfyUI_windows_portable\ComfyUI\comfy\samplers.py", line 334, in calc_cond_batch
output = model.apply_model(input_x, timestep, **c).chunk(batch_chunks)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\comfui\ComfyUI_windows_portable\ComfyUI\comfy\model_base.py", line 191, in apply_model
return comfy.patcher_extension.WrapperExecutor.new_class_executor(
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\comfui\ComfyUI_windows_portable\ComfyUI\comfy\patcher_extension.py", line 113, in execute
return self.original(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\comfui\ComfyUI_windows_portable\ComfyUI\comfy\model_base.py", line 235, in _apply_model
model_output = self.diffusion_model(xc, t, context=context, control=control, transformer_options=transformer_options, **extra_conds)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\comfui\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1778, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\comfui\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1789, in _call_impl
return forward_call(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\comfui\ComfyUI_windows_portable\ComfyUI\comfy\ldm\krea2\model.py", line 225, in forward
return comfy.patcher_extension.WrapperExecutor.new_class_executor(
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\comfui\ComfyUI_windows_portable\ComfyUI\comfy\patcher_extension.py", line 113, in execute
return self.original(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\comfui\ComfyUI_windows_portable\ComfyUI\comfy\ldm\krea2\model.py", line 244, in _forward
context = self._unpack_context(context)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\comfui\ComfyUI_windows_portable\ComfyUI\comfy\ldm\krea2\model.py", line 285, in _unpack_context
raise ValueError(
ValueError: Krea2 expects conditioning with 12x2560=30720 features (a 12-layer Qwen3-VL stack) but got 2560. Load the text encoder with CLIPLoader type 'krea2'.[INFO] Prompt executed in 131.78 secondsUse that guy’s branch above. @m8rr
git clone -b Dynamic-VRAM https://github.com/m8rr/ComfyUI-GGUF#463 (comment) https://github.com/m8rr/ComfyUI-GGUF/tree/Dynamic-VRAM
@RealRebelAI Your GGUF works fine in my environment, though I'm not sure if the calculations are correct. The GGUF from the link provided by @makisekurisu-jp throws a [WARNING] unet unexpected: ['last.down.weight', 'last.up.weight'], but it does work.
And ComfyUI's official BF16 doesn't have ['last.down.weight', 'last.up.weight'], so it seems safe to ignore.
ideogram4, boogu, krea2 / UNet and TE both function properly using GGUF. While Gemma-4 E2B appears to have some problems, it operates well in my specific environment, including vision recognition.
When I switch to using the GGUF model provided by @RealRebelAI, I don’t get the warning:
[WARNING] unet unexpected: ['last.down.weight', 'last.up.weight']
https://huggingface.co/realrebelai/KREA-2_GGUFs[INFO] got prompt [INFO] Model Krea2TEModel_ prepared for dynamic VRAM loading. 5619MB Staged. 0 patches attached. Force pre-loaded 243 weights: 1158 KB. [INFO] gguf qtypes: Q8_0 (263), F32 (166), F16 (1) [INFO] model weight dtype torch.float16, manual cast: None [INFO] model_type FLUX [INFO] Requested to load Krea2 [INFO] Model Krea2 prepared for dynamic VRAM loading. 12989MB Staged. 0 patches attached. Force pre-loaded 160 weights: 2824 KB. [INFO] 0 models unloaded. [INFO] Model WanVAE prepared for dynamic VRAM loading. 241MB Staged. 0 patches attached. Force pre-loaded 60 weights: 61 KB. [INFO] Prompt executed in 94.13 secondsUse that guy’s branch above. @m8rr
git clone -b Dynamic-VRAM https://github.com/m8rr/ComfyUI-GGUFThat fork was last updated 5 months ago. How is that supposed to work with Krea2? And why is there a "Dynamic-VRAM" in the clone command?
Use that guy’s branch above. @m8rr
git clone -b Dynamic-VRAM https://github.com/m8rr/ComfyUI-GGUFThat fork was last updated 5 months ago. How is that supposed to work with Krea2? And why is there a "Dynamic-VRAM" in the clone command?
That command is asking you to clone https://github.com/m8rr/ComfyUI-GGUF/tree/Dynamic-VRAM
Reacted by GlamoramaAttackIn my case, ComfyUI's default settings cause the Text Encoder to reset during heavy tasks like LTX, so I use '--cache-ram 0'. Also, I'm using an NVMe drive, and while adding the '--fast-disk' option mysteriously eats up more RAM for some reason, it does make it faster. Just a heads up!
.\python_embeded\python.exe -s ComfyUI\main.py --windows-standalone-build --fast fp16_accumulation --use-sage-attention --cache-ram 0 --fast-disk
[INFO] Total VRAM 12282 MB, total RAM 32085 MB [INFO] pytorch version: 2.12.1+cu130 [INFO] Device: cuda:0 NVIDIA GeForce RTX 4070 SUPER : cudaMallocAsync [INFO] Model Krea2 prepared for dynamic VRAM loading. 6877MB Staged. 0 patches attached. Force pre-loaded 160 weights: 2824 KB. 100%|████████████████████████████████████████████████████████████████████████████████████| 8/8 [00:10<00:00, 1.31s/it] [INFO] Requested to load WanVAE [INFO] 0 models unloaded. [INFO] Model WanVAE prepared for dynamic VRAM loading. 241MB Staged. 0 patches attached. Force pre-loaded 60 weights: 61 KB. [INFO] Prompt executed in 25.49 seconds [INFO] Model Krea2 prepared for dynamic VRAM loading. 6877MB Staged. 0 patches attached. Force pre-loaded 160 weights: 2824 KB. 100%|████████████████████████████████████████████████████████████████████████████████████| 8/8 [00:12<00:00, 1.53s/it] [INFO] 0 models unloaded. [INFO] Model WanVAE prepared for dynamic VRAM loading. 241MB Staged. 0 patches attached. Force pre-loaded 60 weights: 61 KB. [INFO] Prompt executed in 12.72 seconds [INFO] Model Krea2 prepared for dynamic VRAM loading. 6877MB Staged. 0 patches attached. Force pre-loaded 160 weights: 2824 KB. 100%|████████████████████████████████████████████████████████████████████████████████████| 8/8 [00:12<00:00, 1.54s/it] [INFO] 0 models unloaded. [INFO] Model WanVAE prepared for dynamic VRAM loading. 241MB Staged. 0 patches attached. Force pre-loaded 60 weights: 61 KB. [INFO] Prompt executed in 12.82 secondsReacted by GlamoramaAttackThanks for the infos, @makisekurisu-jp and @m8rr !
Use that guy’s branch above. @m8rr
git clone -b Dynamic-VRAM https://github.com/m8rr/ComfyUI-GGUFThat fork was last updated 5 months ago. How is that supposed to work with Krea2? And why is there a "Dynamic-VRAM" in the clone command?
That command is asking you to clone https://github.com/m8rr/ComfyUI-GGUF/tree/Dynamic-VRAM
Work but got error torchdynamo which solutions is i disabled it detail i share at Reddit stable diffusion
https://www.reddit.com/r/StableDiffusion/comments/1uf66bv/krea_2_turbo_quant_2_bit_on_750_ti_4gb_and_city96/@tukangcode
It seems like the Torch compile exception handling isn't working as expected. It requires installing triton-windows. Check out this: https://github.com/triton-lang/triton-windowsYou should be able to bypass it either by using your method or by setting "set TORCH_COMPILE_DISABLE=1" in run.bat
m8rr@1c113ea
Fixed, I think... so hopefully it works now without having to bypass. Not that it matters for performance, though.Thanks again, I've learned a new command and your GGUF fork, @m8rr, is a big help for me because I don't like the same clip in gguf and safetensors format wasting harddrive space (prefer gguf for ComfyUI and LM Studio).
Root cause found — it's not a missing Krea2 feature, it's a one-line arch gate in `loader.py`. `qwen3vl` is in `TXT_ARCH_LIST`, but the mmproj branch reads `if arch == "qwen2vl":`, so the Qwen3-VL vision tower is never loaded for `qwen3vl` encoders. Without the vision keys, `comfy.sd.detect_te_model()` falls back to `QWEN3_4B`, the KREA2 branch (`clip_type == KREA2 and te_model == QWEN3VL_4B`) never fires, and you get `expects conditioning with 12x2560=30720 features but got 2560`. Fix (loader.py only): add a Qwen3-VL vision map, remap the GGUF DeepStack layer ids (4B: [5,11,17] / 8B-32B: [8,16,24]) to the sequential 0..N-1 that ComfyUI expects, and change the gate to `if arch in ("qwen2vl", "qwen3vl")`. Verified offline against `qwen3vl_4b_fp8_scaled.safetensors` (315/315 keys, cos ≈ 1.000) and on hardware (Intel Arc B580, Krea2 sampling 2.85 s/it). PR incoming.
- added a commit that references this issue
on Sep 24, 2026
diffusion_models:
https://huggingface.co/realrebelai/KREA-2_GGUFs/tree/main/TURBO
https://huggingface.co/realrebelai/KREA-2_GGUFs/tree/main/BASE
text_encoders:
https://huggingface.co/unsloth/Qwen3-VL-4B-Instruct-GGUF