V4 - Easy Krea 2 Workflow +LORA +UPSCALER +PROMPT ENHANCER +I2I +CHECKPOINT MERGER
V4 is V3 with an added CHECKPOINT MERGER Option.
Instructions for CHECKPOINT MERGER are in the workflow
No weird nodes.
Not to many unnecessary options ;)
Just easy and simple
V3 - Easy Krea 2 Workflow +LORA +UPSCALER +PROMPT ENHANCER +IMAGE2IMAGE
V3 is V2 with an added I2I Option.
Instructions for I2I are in the workflow
No weird nodes.
Not to many unnecessary options ;)
Just easy and simple
V2 - Easy Krea 2 Workflow +LORA +UPSCALER +PROMPT ENHANCER
No weird nodes.
Not to many unnecessary options ;)
Just easy and simple.
V1 - Easy Krea 2 Workflow +LORA +UPSCALER
No weird nodes.
No unnecessary options.
Just easy and simple.
Download-Links included in worklflow for the files you need.
Disclaimer:
This workflow was created by taking the best steps from other workflows and make it as easy as possible to use.
Description
This workflow contains my old V1 and the new V2 version.
FAQ
Comments (27)
Why is the VAE model referenced in the workflow different from the Downloads?
KREA2_qwen_image_vae.safetensors
same for Krea2_qwen3vl_4b_bf16.safetensors
Hello, i get three error messages:
-The value Krea2_qwen3vl_4b_bf16.safetensors for Settings's clip_name is not available.
- The value Krea2\Krea2_Turbo_BF16.safetensors for Settings's unet_name is not available.
-The value KREA2_qwen_image_vae.safetensors for Settings's vae_name is not available.
I have used the original workflow and the prescribed models, without any changes.
Have anyone an idea what is my mistake?
Krea2_Turbo_BF16.safetensors, Krea2_Turbo_BF16.safetensors and Krea2_Turbo_BF16.safetensors, thats how i named the files on MY pc. Your files have diffrent names. You have to choose the right ones from the dropdown menu. If you dont't have the right files, i added the download links in the workflow -> The black box on the left.
@Herastura Thanks for your answer. But that was not the problem, i have used the linked models in the black box. The node "settings" doesn't run for me, why ever. Therefore i switched it to a "load clip"- , a "load vae" - and a "load diffusion" - node. The second problem was the clip-model "qwen3vl 8b BF16" generates an error message (problem with the krea2 matrix). Therefore i used the qwen3vl_4b_fp8_scaled.safetensors model. That runs.
@Mikel007 Happy to hear that it works now.
For some reasons, i dont' have that problem.
But i will add the qwen3vl_4b_fp8 in my next release (download link) and add an info.
If there are any other problems with that node, you can just unpacki it and work with the standard nodes: Open the workflow in ComfyUI, right-click on the settings node and then choose "Convert to nodes".
@Herastura Hello, aaah that is new for me, i don't know that you can make a new node from a couple of nodes. Thanks for the tip.
I'd like to try this workflow, but unfortunately it's not working. I've tried everything in the "Settings" node... but it seems to require specific filenames and/or paths.
Hey, you can try to unpack the group-node and see if this solves the problem: Open the workflow in ComfyUI, right-click on the node and the choose "Convert to nodes".
@Herastura Thanks, that worked.
Now I only have one error message:
KSampler (Pass 1)
ValueError: Krea2 expects conditioning with 12x2560=30720 features (a 12-layer Qwen3-VL stack) but got 4096. Load the text encoder with CLIPLoader type 'krea2'.
@romancer_in_mind I think this could be a mismatch of ClipLoader and Checkpoint...but i dont't have that problem for some reasons.
Means, if you use Krea_2_Turbo_FP8, then you should use the FP8 Version of the CLIPLoader: qwen3vl_4b_fp8.
But i only added the link to the BF16 Version (qwen3vl_4b_bf16) of the CLIPLoader.
In short: Try to match the Checkpoints BF-version with the CLIPLoarder BF-Version.
Krea_2_Turbo_FP8 -> qwen3vl_4b_fp8
Krea_2_Turbo_BF16 -> qwen3vl_4b_bf16
You can dowload the missing qwen3vl_4b_fp8 here: https://huggingface.co/Comfy-Org/Krea-2/tree/main/text_encoders
If this doesn't work, please download V1 of my workflow and see if that works.
Link to V1: https://civitai.red/models/2790822/easy-krea-2-workflow-lora-upscaler?modelVersionId=3145017
@Herastura Bullseye!
I'm using the Diffusion Model BF16.
But I had to switch the ClipLoader from qwen3vl_8b_bf16 to qwen3vl_4b_bf16.
Now it works, and I'm excited to see what I can do with your workflow... :-)
Thanks for your support!
@romancer_in_mind Ah yes, this could be a VRAM problem of your GPU.
The 8b_bf16 version loads almost 18GB in your VRAM, while the 4b_bf16 version "only" loads about 9GB in your VRAM.
Her a link to all, or at least to the most used, CLIPLoader for Krea2 if youe ever need to change something: https://huggingface.co/Comfy-Org/Qwen3-VL/tree/main/text_encoders
Hm... my GPU has 32 GB VRAM. 😲
@Herastura With qwen3vl_8b_bf16 I get this error:
KSampler (Pass 1)
ValueError: Krea2 expects conditioning with 12x2560=30720 features (a 12-layer Qwen3-VL stack) but got 4096. Load the text encoder with CLIPLoader type 'krea2'.
ClipLoader type krea2 is set.
@romancer_in_mind I have the same problem and it's NOT a caused by the workflow.
Tried the combination of "Krea 2 Turbo BF16" and "qwen3vl_8b_bf16" with the official krea2 workflow providet in the ComfyUI Templates and i get the same error.
Then i asked AI and they confirmed that the 8b-version is not for Krea2.
Just uploadet a new workflow.file without the 8b.
That work for me thank you ♥
Never seen that error. Is your comfyui up2date?
If V2 doeset work for you, please try V1.
If you really need the Prompt Enhancer, then try the Krea2-Templates providet by ComfyUI.
"Krea-2 Int8: Text to Image" or "Krea-2: Text to Image", they both have a Prompt Enhancer.
@Herastura template comfy work but im interested, you workflow can help me with lora manager
@Herastura Update all is work now thank you
@Fadoo2077 Nice!
Works here. Thank you!
I only have 8 gb vram, can I use the fp8 version with this?
Probably not, bur you can try. If it does not work, you can try it with NVFP4 or GGUF Q4. They use less VRAM.
@Herastura ok. Sorry if that was a n00b question. I'm brand new to krea2. thanks for the response!
❤️❤️❤️