- More cool models.
ℹ️ V2: Recommend to use Dev with strength 1.0-2.0, results with Schnell more volatile - strength 1.4 with 5/6 steps seems to work ok most of the times, but not as good as Dev. If you use sampler with CFG, set CFG=1.
ℹ️ V2: Improved consistency, trained for 53 hours, with multiple resolutions, and dataset of 4100 images that included realistic images, cartoons and illustrations.
ℹ️ V1: I recommend to use Dev with strength 1.0-3.0(not sure about higher values). Results with Schnell don't look as good, use strength 1.0-4.0. If you use sampler with CFG, set CFG=1.
ℹ️ V1: Trained for 25 hours with dataset of 4100 images that included realistic images, cartoons and illustrations.
ℹ️ ComfyUI workflow included in every example image - download the image and drag to your ComfyUI - Civitai seems to ignore it.
Description
ℹ️ ComfyUI has a change related to Flux Loras strength - My recommendation are for ComfyUI versions before it.
ℹ️ Recommend to use Dev with strength 1.0-2.0, results with Schnell more volatile -strength 1.4 with 5/6 steps seems to work ok most of the times, but not as good as Dev. If you use sampler with CFG, set CFG=1.
ℹ️ V2: Improved consistency, trained for 53 hours, with multiple resolutions, and dataset of 4100 images that included realistic images, cartoons and illustrations.
ℹ️ ComfyUI workflow included in every example image - download the image and drag to your ComfyUI - Civitai seems to ignore it.
FAQ
Comments (25)
Is it me or the v2 has worse results than v1 ?
Please make sure that you adjusted the strength(and CFG if you have it) from description of V2.
Also make sure that your ComfyUI is updated.
V2 produced much better and consistent results than V1 in a few hundreds of generated images that I made when I tested it.
Let me know if you discover what might be the problem.
@alexds9 Oh thank you for your reply, i haven't tried yet, But i haven't CFG, i have a flux guidance, is it the same ? The strength was 1.5 on flux DEV
@devher Flux guidance is what I use too, so it's fine. Strength 1.5 with Dev should give you the optimal results.
Please check that you don't have errors in your ComfyUI console - maybe it's not even loading the Lora due problem with downloading the file, it happened to me before.
Sometimes I forget to change steps, when switching models, maybe you have some small issue like that.
@alexds9 My bad, i used the wrong workflow, the strenght was 1... Sorry for the lost time. But thank you for your explainations. I also restarted comfyui and now it seems to be good !
@devher
Thank you for letting me know, I am happy to hear that it is working for you.
could you please add your prompts to the example images? not everyone uses comfy ui.
Sorry, I generated hundreds of images with wildcards, it's not practically for me to do.
There is nothing special about those prompts, they have description of naked woman, style and place. So you can easily make your own prompts, or Civbot can help you, or some free online LLM.
There are sites that can open ComfyUI workflows, but I have no idea if you will be able to load the workflow and see the prompt...
@alexds9 I don't understand why it's so hard to answer the question that probably is being made by A LOT of people, you know, simply saying "Sure, here is one of my example prompts: blablablabla". Are they copyrighted or something? 😄
@WetnWild
I already explained why.
ComfyUI is complicating the process, now I'm using Forge so it's not a problem.
But I never got to make another version for this model because the training script had a bug.
I get the following error trying to use your LoRA:
Error occurred when executing CR Apply LoRA
Stack: Error while deserializing header: HeaderTooLarge
File "/workspace/ComfyUI/execution.py", line 316, in execute output_data, output_ui, has_subgraph = get_output_data(obj, input_data_all, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb)
And it continues... maybe a conflict with some custom node? Any idea how to fix it?
I am using my workflow "Tenofas FLUX v.2.0"...
I would recommend to update your ComfyUI. With updated ComfyUI and workflow that can load Flux Loras - it should work.
And maybe you are using node that doesn't support Flux Lora, I am just guessing...
@alexds9 I just updated everything, still the same error... I am using CR LoRA Stack node, and with some other FLUX LoRA's it works... maybe I should try with a different Lora Loader node.
@Tenofas
Yes, there are a few variations of Loras that ComfyUI introduced after a few days.
Your node probably does not fully support all Lora types yet.
If you can try with regular node that loads Loras - it should work fine.
@alexds9 Yes, tested with another workflow, a more simple one, and it works. I changed the Lora Loader in my larger workflow, and still get the same error... probably it's some other node that has trouble... will test every single node. Thank you!
@Tenofas
Thank you for the update, good to hear that it started working for you.
First FLUX-Pussy on whole Civitai.
I'm sure it wouldn't be the last...
@alexds9 0 to 1 is amazing and interesting, 1 to 100 just repeating and boring.
@Aseer
Maybe, we still need to establish the 1 part first...
@alexds9 of course, but, after all, you made great job. Thanks for you contribute.
@Aseer
Thank you very much.
Using this lora with Dev, euler beta, CFG=1 and Guidance=3.0 I always get the cartoonish look while looking for realism, what I am doing wrong?
Maybe you have a complex prompt that pushes the model towards a cartoon style. Start with a short prompt and add terms like "realistic" and similar descriptors. Most of the training was on realistic images, and I haven't noticed a bias towards cartoon style during my testing.
I've actually wanted to create a newer version of this model for a while, to achieve a lower default strength and potentially improve the results. But, a critical bug in the training script was preventing me from doing so. Fortunately, the developer recently fixed the issue after I’d been crying about it for weeks. I've been busy with other things, but I plan to revisit this model this week and start new training.
Details
Files
Flux_Improved_Female_Nudity_v2.safetensors
Mirrors
Flux_Improved_Female_Nudity_v2.safetensors
Flux_Improved_Female_Nudity_v2.safetensors
Flux_Improved_Female_Nudity_v2.safetensors
Flux_Improved_Female_Nudity_v2.safetensors
101-Flux_Improved_Female_Nudity.safetensors
Flux_Improved_Female_Nudity_v2.safetensors
Flux_Improved_Female_Nudity_v2.safetensors
flux1dev_nudity_female_v2.safetensors
Flux_Improved_Female_Nudity_v2.safetensors
Flux_Improved_Female_Nudity_v2.safetensors
Available On (2 platforms)
Same model published on other platforms. May have additional downloads or version variants.