Trigger words : kyudo-aiming,aiming,arrow (projectile),bow (weapon),drawing bow,holding arrow,holding bow (weapon),partially fingerless gloves,yugake
Qwen-v1.0 21/02/2026 added
Trigger word ? : kyudo-aiming
CFG : 1.5-3.0
AuraFlow shift : 2.0-3.0
Flux-v1.0 04/05/2025 added
Strength : 0.8-1.0
The Drawing accuracy has been improved from the beta version. But, unintended drawing may occur.
NAI-ePred-v1.0 22/03/2025 added
Trigger words : kyudo-aiming,holding bow \(weapon\),drawing bow,bow_string,arrow \(projectile\),aiming,partially fingerless gloves,yugake
Strength : 0.7-1.0
The drawing accuracy and stability may be improved over the previous versions.
Pony-v3.0 02/03/2025 updated
Trigger words : kyudo-aiming,holding bow \(weapon\),drawing bow,bow_string,arrow \(projectile\),aiming,partially fingerless gloves,yugake
Strength : 0.7-1.0
The image quality may be slightly improved over Pony-v2.0. But, the drawing accuracy and stability may be slightly worse.
IL-v1.0 03/12/2024 added
Strength : 0.7-1.0
The drawing accuracy may be roughly the same as Pony-v2.0.
Flux-beta 19/10/2024 added
The drawing accuracy has improved, but there are still some problems.
1. The background is blurred.
2. Sometimes the drawing is not correct.
Pony-v2.0 28/06/2024 updated
The training data were refreshed.
Strength : 0.7-1.0
Pony-v1.5 15/06/2024 updated
Strength : 0.6-0.9
It may have improved the accuracy of drawing the bow, arrow and bow-drawing motion.
SD1.5-v1.0 14/06/2024 added
Strength 0.6-1.0
The version tends to draw similar to v1.0 (pony).
The drawing is fast, but unstable.
SDXL-v1.0 13/06/2024 added
Strength : 1.0
The accuracy of drawing the bow, arrow and bow-drawing motion is slightly lower than the pony version.
Recommend output at 1:1 aspect ratio.（Any other aspect ratio may result in unstable rendering.）
Pony-v1.0, 1.2
Strength : 0.6-1.0
Increasing the strength improves the accuracy of drawing the bow, arrow, and bow-drawing motion, but may adversely affect image generation.
Description
FAQ
Comments (5)
The pony version of this lora is really good. I also made a SD1.5 Kyudo LoRA 4 months ago, and I want to upgrade it to a Pony LoRA. Unfortunately, all the attempting for pony upgrade has been failed. Could you share your training parameters?
Thanks for your comment. I too did not expect to be able to create a model that could be rendered so well reproducibly. As for the key settings, I think "LR Scheduler"="cosine_with_restarts","Optimizer="Lion" and "LR warmup (% of total steps)"="10". In the SD1.5 version, in addition to the previous setting, the drawing was improved by setting "Network Rank (Dimension)"="64" and "Network Alpha"="32".
@hayake5 thank you for your reply，and how do you set the learning_rate and doyou use the basic pony model as the base model for training
@Oraculum The learning_rate were 0.0001 (1e-04) at v1.0 and v1.2, and 0.00005 (5e-05) at v1.5. The base model were "real pony real_anime_No.04" at v1.0 and "Pony Diffusion V6 XL" at v1.2 and v1.5.
@hayake5 thank you. i will try again