NOTE :Base model : Z Image Turbo & Flux1.
Showcase images were created with ComfyUI and are in PNG, include metadata & workflow.
Description
Model trained on civitai with 26 own images.
Recommended resources : Fluxmania
Recommended settings : dpmpp_2m or heunpp2 / sgm_uniform / 25 steps / guidance : 3.5 (if you use a different model than the one I used, adjust the guidance)
Weighting : around 0.6
FAQ
Comments (15)
This is one of my favourite LoRA's, I highly recommend it.
Thanks for your feedback
Cant run this LORA in forge UI, dunno why but the LORA is not working.
Any clues?
Sorry, I'm no help on Forge, I don't use that interface. But I know that many on Forge use my models without any problems.
已下载学习 感谢 标记
This style has nothing to do with anime.
What is this style called then? Enlighten us.
@psychananaz No anime style? Totally anime style? Depends on the wind direction, the tide schedule, and whether my cat sneezes at the right moment. In short: Schrödinger’s anime.
@Adel_AI perchance
@psychananaz Google it. You can do it, I believe in you.
@magenta the ratio is really on your side hombre
I used to say that. I have the original DVDs of Akira and GiTS here. Akira, particularly, was before the Disneyization with huge eyes. It was a cel shaded 'cartoon' style lacking anything but the barest hint of shading, except for some background stills. They still fit the style though.
Fast forward to Civitai, today, and it appears the word "anime" means: Not a photograph. 😉🤣 Yes, if you're a pro, this can chafe, as clients can be very picky. BUT... if you get too caught up in correctness, you miss half the fun. There are many LoRA out there that can do some really interesting things to your image that have little or NOTHING to do with the title. And since some of the LoRA sample images seem to have nothing to do with the title (and the maker might not speak English or even use our character set), you gotta loosen up a little. 💃 It's a pixel party. 🎉
Hello Adel, small hiccup, im unable to run this amazing lora with flux nunchaku. I keep getting this error - 'text_encoder.text_model.encoder.layers.6.self_attn.v_proj.weight'.
Weird thing is, it runs without issues with the fp8 model. I am able to make use of other Flux loras with nunchaku, only your ones fail with this error. Do you have any idea or workaround for this? Please respond if you get time, as I love to use your loras :)
Hey;
This LoRA was trained (on Civitai) on a different text encoder structure than the one included in Flux Nunchaku. Specifically, the model is looking for a weight in the 6th attention block (v_proj) that doesn’t exist in Nunchaku’s current text encoder.
I'm not a nunchaku user, so I don't really have any tips to offer you to get around this error.
Adel_AI Aww, that sucks. Anyway, thanks for the note. I will see if I can find a way to port these lora's somehow, if it's even possible.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.