KonanMix Illustrious XL is a merge of Illustrious-XL-v0.1.
This model is intended to be used with ressources trained on Illustrious-XL-v0.1. Other SDXL or Pony ressources may not work.
This model goal is to get a better anatomy and style control without affecting the quality and LoRA compatibility.
To help you start with Illustrious prompting here's my recommended postive/negative prompts:
Positive: best quality, amazing quality, very aesthetic, absurdres
Negative: multiple views, sound effects, lowres, (bad), text, error, fewer, extra, missing, worst quality, jpeg artifacts, low quality, watermark, unfinished, displeasing, oldest, early, chromatic aberration, signature, extra digits, artistic error, username, scan
Merge parameters:
SDXL DPO with merge ratio = 0.5
Custom Style* with merge ratio = 0.7
Merge and Save precision = bf16
*Custom style LoRA is trained on 850 pictures with various style from Houtengeki artist.
Illustrious XL has been developed with resources provided by OnomaAI
For detailed information, please refer to Illustrious-XL official release page.
Description
FAQ
Comments (21)
I don't have much thoughts on Illustrious yet, but I appreciate your effort on making Frieren dial down on the booze 👍
Try it. For anime it's better than Pony.
Prompting is very different and more like SD1.5 style because it's trained on Booru caption.
@Konan Isn't pony also trained on Booru captions? All of the tags I used before seem to work with Illustrious too. Some even better, some worse because it mostly does anime.
@blaqua True but Pony is trained with particular caption :
- scoring
- short descriptive sentence
- booru captions
- hashing artist name and some characters
Illustrious is trained directly on Booru tags. So same prompting than SD1.5 and lot of characters and artist names can be use without any LoRA.
Have you tried SPO instead of DPO? Apparently it's supposed to be better, i've been using the lora version, strength slightly above 1
I didn't. Thanks for the info. I will try it later.
i also use the lora, what strength do you usually usee?
@low_channel_1503 above 1 a little, too high and it bleeds.
Please upload the workflow for the merging method
To do what? All is already written.
Very good art style. It would be perfect if the anatomy could get improved.
Hi, could you help me with something? I would really like to create my own checkpoint. But I honestly don't know how to do it. How do you create them?
You have various methods and tools, I did a merge with Kohya_ss.
@Konan do you remember any tutorial on youtube how to create a checkpoint using this Kohya_ss.?
@ApexThunder self learned
Can anyone tell me what the difference between IL and Pony?
Both SDXL checkpoints but they are too different so LoRA trained on one will not work with the other.
Illustrious has better compatibility with other SDXL ressources though. Pony is more like his own ecosystem.
@Konan not true, or more like it depends on the checkpoint, used pony loras on an IL-checkpoint and it just worked. -> https://civitai.com/posts/8633366
Maybe you can't use IL loras on pony checkpoints, but it works at least for the other checkpoint.
hey, model is not available for onsite generation anymore. Can you please enable it?
Yo, I can only get 2 models hosted for onsite generation.
Overall I think my NoobMix is outperforming this model and it's available for onsite generation.
https://civitai.com/models/895531/konanmix-noobai-xl-illustriousxl
more ver plzplzplzplzplzplz
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.