Update: Following a request, i reposted the old versions "IXL Next" and "IXL V0" here as previous versions of Reborn.

Usage

• Steps 25-30 should be enough, 40 doesn't hurts :D

• CFG: i tested with 5, going lower may be fine

• Scheduler Sampler: Euler A and DPM++ 2M should be fine

• Clip Skip: the usual 2

• Adetailer: recommended :D

• Negative prompt: IllusN3g, deep_negative_pony (for a more authentic feeling :D)

Adetailer trick: To avoid the "box effect", what i usually do is set this in the inpaint section:

Long Story™

One thing has been frustrating me for a long time. Neither the Next or V0 version of BancinIXL felt close enough to BancinXL. So, i deleted those got back to work.

Here, i started back from scratch to try and get a closer feeling, using two different approach using AnBan 1.0:

• Using supermerger with "Add Difference" in "Train difference" using BancinXL-Pony (leveraging xyz plot, i got that 0.5 was the max i could incorporate this way)

• Directly finetuning the checkpoint using the dataset (200 pictures) i previously used for a BancinXL LoRA (the method was the same used for HoJ, but since the dataset was small, it got a strong bias)

This left me with two differents checkpoints that i merged together with an intermediate step for each (closer to merge, closer to finetune), carefully selecting the block pairs that i liked.

Feel free to experiment, this is still not a "Pony like illustrious", but it should feel more western/comic style that AlterBan or Quirk 😅