This model was finetuned with custom scripts, the goal was to avoid bleeding by focusing only on the core features, without altering the base model sharpness, lighting, style and preserve faces, so you can use a character lora without confusing the weights.
Initially made for my Blender addon https://pgcrt.github.io/
Simple workflow with prerendered prompts (CRT Nodes)
Download the image and drop it into ComfyUI
https://civarchive.com/images/124361359
Description
FAQ
Comments (23)
The body shapes looks great. Will def try it!
Last time I've tried finetuning flux klein it was pretty long to train. Do you train with 1024px images or smaller?
Yes it's long, I trained up to 1024px buckets, but images longer side max is 1536px
Stop releasing lore already, I'm running out of disk space))
I checked your OpenBlender video. This is insane, amazing work! 🙌
Thanks! This one? https://www.youtube.com/watch?v=LdsYLxJ3WCc
Yeah this was the release video of the addon, I'm still improving it, it's a nice tool ,)
Can you display or share your workflow and the prompt for example images?
I cannot achieve the effect of your example images using images generated at different resolutions I try. I want to see where the problem lies, whether it's with the process or the prompt.
Download this image and drop it into comfyUI, download / update nodes
https://civitai.com/images/124361359
Prompts are prerendered
For some reason ComfyUI does not recognize its type and gives me an error: Could not detect model type.
So you downvoted the model because your download was interrupted? You are blaming the model for that?
Could you please dial it back a bit and keep things respectful?
Like I said in my comment, I genuinely don’t know the reason behind the error. “For some reason” was meant literally. From my side, there’s no indication that the download was interrupted, so I’m not sure how I would’ve known that. Also curious to know if you can see that from your end and confirm?
If that is the case, I’m more than happy to redownload the model and reverse the downvote after testing it again.
Just to follow up, I’ve now redownloaded the model twice (three times in total), so unless I’m impressively unlucky, corruption probably isn’t the issue.
The “Could not detect model type” error is still persistent, and I also tried it with the provided workflow, same result. So no luck on my end so far.
I’m sure there might be something obvious I'm missing, so I’m happy to take a look if you have suggestions. But for now, it’s still a no-go on my side.
@brownbagel0 I too have the same issue, "RuntimeError: ERROR: Could not detect model type of: C:\ComfyUI_02-27-26\ComfyUI_windows_portable\ComfyUI\models\diffusion_models\: secretSAUCEFlux2klein_v10.safetensors I will attempt a redownload as well and report back, I've had issue like this with Dark Beast models too.
@brownbagel0 I just re-downloaded and tested "the Sauce" got the same error so I downloaded "SNOFS (Sex, Nudes, and Other Fun Stuff) Flux 2 Klein 9b Base & Distilled" distilled version, no issues, works like a charm. I was thinking try another model download...well that isn't it. To bad, "the Sauce" looks interesting, I was looking forward to mess around with it. Ok, brownbagel0 make some room, we are both in the same boat.....hey it could be one of many thing....I'll just move on.
@brownbagel0 I think I figured it out, I'm missing or I'm having an issue with "Sage Attention" I think that what its called, I ran pgc's Comfyui setup, updated / installed everything missing, there was plenty, and I ran it, it got stuck on sage attn then bam that error pops up. On my end problem solved. Did I get it going, no, and I don't care too.....I'm not crazy about Comfyui TBH, everybody has their own twist and configs to it....you can have it. I just figure I'd share my findings, maybe....just maybe it'll help out someone else. Later, Mr. J
@mrj622592 That's for the follow up!
I'm not currently utilizing Sage. Maybe I should. :)
@brownbagel0 That's probably the issue, missing Sage Attention.. I'm just going to go back to Z-Image Turbo for a while anyway. Good luck!!!
I will just add this for visibility, need it to work in forge, fix it with the python script...
https://github.com/Comfy-Org/ComfyUI/issues/12622
model will work in forge and comfy ui after
This is the first real fine-tuned version of Klein 9B that I’ve been able to try, and it pleasantly surprised me. It can be used with character LoRAs without changing their features. The only issue is that it tends to deform fingers and small details more than the normal version. Why don’t you upload the FP16 version?
This model is likely very underrated. Amazing job with this!
This is All in one? Or do I need to add the Vae and the clip?
Please include a description. These are literally 3 lines of text:
Sampler
steps
cfg
Your drop image method requires downloading customnode. Without this, the values in the sampler are not visible
Can you upload a bf16 version? That would be really great!