This is a GGUF quantized version of FLUX.2-klein-4B.
unsloth/FLUX.2-klein-4B-GGUF uses Unsloth Dynamic 2.0 methodology for SOTA performance.
Important layers are upcasted to higher precision.
Uses tooling from ComfyUI-GGUF by city96.
Description
FAQ
Comments (17)
Do you know why is error like FluxParams.__init__() missing 1 required positional argument: 'vec_in_dim' appearing? :/ its both for safetensor and gguf models when it comes to flux2 and i never found answer to the problem. If anyone knows something please let me know.
Are you using comfyui?
@ccollins Yes
@AltairTheArc make sure your running the most recent version and that any nodes you have are also running their most recent version.
@ccollins mm...i'm using 0.4.76 which is from Sep 30, 2025
@AltairTheArc That is out of date which is probably the issue. The current desktop install version is 0.8.23
@AltairTheArc Did you have any luck with updating?
@ccollins nope, couldn't update, got errors, probably driver related if to trust github, and since me gpu vintage and i cant upgrade driver, welp, can't help it. o7 ty for your help.
copy the workflow off the following image and paste it into comfyui:
https://civitai.com/images/121710803
just click on the comfyui: 5 nodes entry on civitai, then paste it into comfyui
@ccollins I did that 1st try, (i mean i did this before you told me this now? like i did what you said on my own before you said it, bad english) I did copied your workflow, it did not worked, so i dissected it, it still did not worked so i got template workflow from comfyui for appropriate model, it did not worked either. PS: I also installed the required nodes.
So far, flux2, is the only model, that i touched, that i cant get it to work.
I succeded with sd1.0 sd1.5 sdxl, pony, noobai, illustrious, wan2.1 wan2.2, flux1, chroma, anima, sound models, voice models, text models, animation models, safetensor and gguf, like, what im trying to say, i know what im doing (i do not, who am i kiding) but i still can not get flux2 to work, be it safetensor, or gguf, and i have like the newest 18.2 comfyui portable at the moment. So i trully have no idea. I get all those Ais for like, knowledge? experience, i dont have much use for them, flux2 liekly is to heavy to get any value out of it, but it rubs me wrong way that i cant get it to work. i will probably because of it do try 3, someday, but well, yeah, ye.
I mean
Thank you for your help <3
Got myself portable version working, so i could load the model, however i got now this error: RuntimeError: mat1 and mat2 shapes cannot be multiplied (512x1024 and 7680x3072) - Eh, i was using Qwen3-0.6b.safetensors loaded into normal Clip loader set to Flux2 device.
Do i have to use .gguf clip? I will rant but honestly, those flux models are more trouble than they probably worth. Ehhh. Anima droped out, i got it working in 3 hours, Flux2 exsists, months passes, and yet i have zero successfully working Fluxes2.
If that helps anyone here is error log:
File "I:\zStableDifusionComfyPortable\ComfyUI\execution.py", line 525, 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 "I:\zStableDifusionComfyPortable\ComfyUI\execution.py", line 334, in get_output_data return_values = await asyncmap_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 "I:\zStableDifusionComfyPortable\ComfyUI\execution.py", line 308, in asyncmap_node_over_list await process_inputs(input_dict, i) File "I:\zStableDifusionComfyPortable\ComfyUI\execution.py", line 296, in process_inputs result = f(**inputs) File "I:\zStableDifusionComfyPortable\ComfyUI\comfy_api\internal\__init__.py", line 149, in wrapped_func return method(locked_class, **inputs) File "I:\zStableDifusionComfyPortable\ComfyUI\comfy_api\latest\_io.py", line 1764, in EXECUTE_NORMALIZED to_return = cls.execute(*args, **kwargs) File "I:\zStableDifusionComfyPortable\ComfyUI\comfy_extras\nodes_custom_sampler.py", line 963, in execute samples = guider.sample(noise.generate_noise(latent), latent_image, sampler, sigmas, denoise_mask=noise_mask, callback=callback, disable_pbar=disable_pbar, seed=noise.seed) File "I:\zStableDifusionComfyPortable\ComfyUI\comfy\samplers.py", line 1052, in sample output = executor.execute(noise, latent_image, sampler, sigmas, denoise_mask, callback, disable_pbar, seed, latent_shapes=latent_shapes) File "I:\zStableDifusionComfyPortable\ComfyUI\comfy\patcher_extension.py", line 112, in execute return self.original(*args, **kwargs) ~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^ File "I:\zStableDifusionComfyPortable\ComfyUI\comfy\samplers.py", line 995, in outer_sample output = self.inner_sample(noise, latent_image, device, sampler, sigmas, denoise_mask, callback, disable_pbar, seed, latent_shapes=latent_shapes) File "I:\zStableDifusionComfyPortable\ComfyUI\comfy\samplers.py", line 981, in inner_sample samples = executor.execute(self, sigmas, extra_args, callback, noise, latent_image, denoise_mask, disable_pbar) File "I:\zStableDifusionComfyPortable\ComfyUI\comfy\patcher_extension.py", line 112, in execute return self.original(*args, kwargs) ~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^ File "I:\zStableDifusionComfyPortable\ComfyUI\comfy\samplers.py", line 751, in sample samples = self.sampler_function(model_k, noise, sigmas, extra_args=extra_args, callback=k_callback, disable=disable_pbar, self.extra_options) File "I:\zStableDifusionComfyPortable\python_embeded\Lib\site-packages\torch\utils\_contextlib.py", line 124, in decorate_context return func(*args, **kwargs) File "I:\zStableDifusionComfyPortable\ComfyUI\comfy\k_diffusion\sampling.py", line 205, in sample_euler denoised = model(x, sigma_hat s_in, *extra_args) File "I:\zStableDifusionComfyPortable\ComfyUI\comfy\samplers.py", line 400, in call out = self.inner_model(x, sigma, model_options=model_options, seed=seed) File "I:\zStableDifusionComfyPortable\ComfyUI\comfy\samplers.py", line 954, in call return self.outer_predict_noise(*args, **kwargs) ~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^ File "I:\zStableDifusionComfyPortable\ComfyUI\comfy\samplers.py", line 961, in outer_predict_noise ).execute(x, timestep, model_options, seed) ~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "I:\zStableDifusionComfyPortable\ComfyUI\comfy\patcher_extension.py", line 112, in execute return self.original(*args, **kwargs) ~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^ File "I:\zStableDifusionComfyPortable\ComfyUI\comfy\samplers.py", line 964, 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 "I:\zStableDifusionComfyPortable\ComfyUI\comfy\samplers.py", line 380, in sampling_function out = calc_cond_batch(model, conds, x, timestep, model_options) File "I:\zStableDifusionComfyPortable\ComfyUI\comfy\samplers.py", line 205, in calc_cond_batch return calccond_batch_outer(model, conds, x_in, timestep, model_options) File "I:\zStableDifusionComfyPortable\ComfyUI\comfy\samplers.py", line 213, in calccond_batch_outer return executor.execute(model, conds, x_in, timestep, model_options) ~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "I:\zStableDifusionComfyPortable\ComfyUI\comfy\patcher_extension.py", line 112, in execute return self.original(*args, kwargs) ~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^ File "I:\zStableDifusionComfyPortable\ComfyUI\comfy\samplers.py", line 325, in calccond_batch output = model.apply_model(input_x, timestep_, c).chunk(batch_chunks) ~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^ File "I:\zStableDifusionComfyPortable\ComfyUI\comfy\model_base.py", line 171, 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 "I:\zStableDifusionComfyPortable\ComfyUI\comfy\patcher_extension.py", line 112, in execute return self.original(*args, kwargs) ~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^ File "I:\zStableDifusionComfyPortable\ComfyUI\comfy\model_base.py", line 210, in applymodel model_output = self.diffusion_model(xc, t, context=context, control=control, transformer_options=transformer_options, extra_conds) File "I:\zStableDifusionComfyPortable\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1776, in wrappedcall_impl return self._call_impl(*args, **kwargs) ~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^ File "I:\zStableDifusionComfyPortable\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1787, in callimpl return forward_call(*args, kwargs) File "I:\zStableDifusionComfyPortable\ComfyUI\comfy\ldm\flux\model.py", line 345, 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, y, guidance, ref_latents, control, transformer_options, kwargs) ~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "I:\zStableDifusionComfyPortable\ComfyUI\comfy\patcher_extension.py", line 112, in execute return self.original(*args, **kwargs) ~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^ File "I:\zStableDifusionComfyPortable\ComfyUI\comfy\ldm\flux\model.py", line 406, in forward out = self.forwardorig(img, img_ids, context, txt_ids, timestep, y, guidance, control, timestep_zero_index=timestep_zero_index, transformer_options=transformer_options, attn_mask=kwargs.get("attention_mask", None)) File "I:\zStableDifusionComfyPortable\ComfyUI\comfy\ldm\flux\model.py", line 182, in forward_orig txt = self.txt_in(txt) File "I:\zStableDifusionComfyPortable\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1776, in wrappedcall_impl return self._call_impl(*args, **kwargs) ~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^ File "I:\zStableDifusionComfyPortable\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1787, in callimpl return forward_call(*args, **kwargs) File "I:\zStableDifusionComfyPortable\ComfyUI\comfy\ops.py", line 392, in forward return self.forward_comfy_cast_weights(*args, **kwargs) ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^ File "I:\zStableDifusionComfyPortable\ComfyUI\custom_nodes\ComfyUI-GGUF-main\ops.py", line 215, in forward_comfy_cast_weights out = self.forward_ggml_cast_weights(input, args, *kwargs) File "I:\zStableDifusionComfyPortable\ComfyUI\custom_nodes\ComfyUI-GGUF-main\ops.py", line 244, in forward_ggml_cast_weights return torch.nn.functional.linear(input, weight, bias) ~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^
use the qwen3_4b.safetensor . I didn't have an issue with it,just read what was needed and downloaded and put in right folder and works.Check templates for requirements. If you wanna use gguf,then replace the checkpoint node with the gguf one and should be fine.Not all text encoders work with everything sadly.
Nope, still same error even with GGUF CLIP oh dear loooooord i hate flux
Ed: Qwen3-0.6b.gguf
it works fine. 0 errors.
Nice to have these checkpoints here. Thanks!
Have you considered adding also the 9B variants? I'm using the 9B Q4_K_M personally, but no one seems to have uploaded it.
Thank you so much for your work!
I also really appreciate that you released the model with weights in FP32 format. For older GPUs, an FP32 model with such a compact 4 GB size is a real lifesaver – it avoids any speed loss and is an absolute king of fast generation.
I really like this distilled model – it handles complex prompts for 4 images quite well, and the generation quality pleasantly surprised me compared to other models. Even though it's the smaller version in the series, I'm very satisfied with the results.
I'm using the GGUF Q8_0 version, and the quality remains excellent even when converting anime-style images into realistic ones – the output is crisp and clear at Full HD resolution.
The background removal is outstanding – I'm amazed that this model can do it so well in just 4 steps.
I should have tried this model sooner – now I've found a new "big brother" among models that will replace my Flux 1 Kontext.