Reakaaka's enhancer [Pony v7]

reakaakaskyreakaakasky
20
139
style
v0.1

Details

Download Files

Model description

Enhancement bundle for Pony v7

Cover images are raw output from the model, default 1MP resolution, no magical plugins, no upscale, no hands/faces inpainting fixes, no negative prompt.

Test run. v0.1 was trained with limited dataset and steps.

This page will be rewritten later (maybe).

Basically same dataset and training settings when I trained the Chroma HD version. So if you want to know more about the dataset and effect, see this page: /model/2014953

Tldr:

For anime and 2D art style:

  • I added a default anime style with simple and straight forward trigger words "anime illustration". Because I can't control the style by only using prompt. Those "style cluster" things are mysterious. At least now I won't get a horrible image without using the "style_cluster_xxx". In theory, the "style cluster things" should still work, as well as other style LoRAs (assume there will be...).

  • Also included a sub dataset full of anime hands. Maybe better hands.

For general realistic image: (?) not tested on Pony v7, copied from Chroma:

  • Better natural texture and details. Better lighting, a little bit cinematic feelings.

  • Less overexposure and oversaturation.

Other thoughts:

  • Copy pasta training settings is not a good idea. The learning rate seems too high.

  • From perspective of an anime trainer. Anime knowledge in pony v7 is very limited. So I will need additional LoRAs (style, characters etc.). In this case, why would I choose pony. I can train those LoRAs on chroma or even qwen, and get better results.

  • Training on Chroma HD only needs 1.5x computational time than AuraFlow. But the quality of Chroma HD is much better because the base model is better. Just feel training on AuraFlow is not worth the cost.

  • From perspective of a user, AuraFlow a big model and needs lots of steps. Even slower than qwen (with lightning LoRA and low steps). E.g. On my setup, AuraFlow needs 30 steps to get good quality and this takes 20s. But Qwen only needs 8 steps (with lightning LoRA) and only takes 16s, and the quality is on the next level. Why would I use AuraFlow?

  • Pony v7 is a good model. Sadly other models are just better.

  • Anyways, at least I tried.

License

This model is released under Apache License 2.0.

Additional terms: Selling or monetizing models that merged this LoRA is prohibited.

Update logs

  • (11/5/2025) v0.1 init release.

You have read this far, don't forget to leave a feedback in discussion. Don't write a review, Civitai review system is so hard to find and, most importantly, read...

Images made by this model

No Images Found.