AnyLoRA
Add a ❤️ to receive future updates.
Do you like what I do? Consider supporting me on Patreon 🅿️ or feel free to buy me a coffee ☕
For LCM read the version description.
Available on the following websites with GPU acceleration:
Remember to use the pruned version when training (less vram needed).
Also this is mostly for training on anime, drawings and cartoon.
I made this model to ensure my future LoRA training is compatible with newer models, plus to get a model with a style neutral enough to get accurate styles with any style LoRA. Training on this model is much more effective compared to NAI, so at the end you might want to adjust the weight or offset (I suspect that's because NAI is now much diluted in newer models). I usually find good results at 0.65 weigth that I later offset to 1 (very easy to do with ComfyUI).
This is good for inference (again, especially with styles) even if I made it mainly for training. It ended up being super good for generating pics and it's now my go-to anime model. It also eats very little vram.
Get the pruned versions for training, as they consume less VRAM.
Make sure you use CLIP skip 2 and booru style tags when training.
Remember to use a good vae when generating, or images wil look desaturated. Or just use the baked vae versions.
Description
Use this for generating images, not for training.
FAQ
Comments (21)
Did I understand correctly that this model is well suited for creating lora and other things? Sorry, my English is not very good :(
How this works?
it's just a normal model. Use it as base when training loras. Or use it with lora made by others to generate perfect style images and accurate characters
yo my favorite creator uploading another model, i have question man, so correct me if im wrong, you suggest this model to train any character right? i follow your guide on creating hybrid character for lora that could transcent between realistic to semi realistic/anime style, you suggest to train in in your other model NeverEndingDream, and it works like magic, will using this also achieve the same goal?
this can be used for the opposite. You can train anime characters and make them realistic, which is what I do all the time (see my Nico Robin LoRA or my Ganyu LoRA). This is especially true if you add photos to the dataset.
Keep in mind that the tag "photorealistic" also has a big effect on this model.
keep in mind this is also great on its own. I'm uploading some examples in a bit.
@Lykon dayuuum, that sicks, will try for sure
Sorry to jump in here. Could I get a link to the guide @basskubetot mentioned? Thank you!
I will try it for sure!
Ill certainly try it, thank you very much, you are amazign!
Ok I'm dumb.. What makes this specific model better to train LORA's with? Can you give me some examples or perhaps more information on this?
It's neutral enough and contains various modern models, meaning the changes your lora contains will be applied to those models with more accuracy. I wanted to post a new lora but I'll wait for civitai to fix this new nsfw thing first....
What a mega mix! Great
I express my admiration for your proposal to encourage more people to create LORA.
Looks good, ill see if it improves my loRA's, maybe it will inspire me to make my own checkpoint for making loRA's
so ... is it beter/on par to use this merge model for training instead of NAI/AV3 ?
Kohya ss and dreambooth alike ?
what makes a merge model good for training ?
thx for any help ;)
a3 was never that good for training and I always had better results with this compared to NAI.
A merge that's good for training is one without a distinctive style and with lots of range, plus it should to be as small as possible. However that depends on what you want to train. See my Jack Sparrow LoRA for a totally different example.
@Lykon thx a lot for this clear and complete answer. the jack sparrow example is a great insight.
@dasasdas nai trained lora don't really work too well on aom3, basilmix (so chilloutmix), and in general new models which have a small nai percentage. I personally had more luck training on this. Could be me, but I still wanted to share