WARNING!
This LoRA model is trained on a real person's likeness and is intended solely for creative use. It must not be used for any harmful or malicious activities directed toward any individual. The creator of this model assumes no responsibility for any misuse of the LoRA for harmful purposes.
If any misuse is reported or a takedown request is submitted, the LoRA will be promptly removed.
FAQ
why Pony instead of SDXL?
Personally, i prefer Pony model because it achieved better anatomy and allow for more creativity in image generation. Hence, i focused the training on pony model. There are no intention to train on SDXL model, but FLUX model is possible in future when it is more matured and allow for possibility to training using low end PC.
What is the standard of Pony model that you choose?
Some might wonder why am I keep retraining the same lora in different model again and again. Why not use one standard model (i.e Pony v6 / PonyRealism). Reason being that the resemblance work best in the said model.
Myself is constantly testing various checkpoints and I select the checkpoint I trained based on the standard of realism it achieved while able to maintain the freedom of Pony prompting.
Hence, currently I am focused on using only_porn checkpoint (despite the nsfw name, it is still capable of generating PG photo), currently from my testing, it achieve good realism in pony standard in terms of skin and saturation while essentially can generate the same image as other 2.5D model. The other option that does similar effect is 3010xx-nc but it leans more towards 2.5D instead of realism.
Any intentions to train on ponyrealism as it is the one of the more famous realism model?
as of current state, no. Ponyrealism just doesn't click with me as the reason mentioned in the previous question.
Do you accept any character training request?
I do intend to open Patreon / Ko-fi in future for commissions. As of current state, I simply train on whichever character I am interested in.
Latest upload are trained in DORA, do I need any additional extension to use it?
from my testing with Reforge webUI & Forge webUI, no. You can use DORA essentially same as LORA without any changes while it giving better result, this is the main reason i started training on DORA.
Installing the webui
I am using reforge webui as my primary sd generation tools.
You can install following instruction in the hub: https://github.com/Panchovix/stable-diffusion-webui-reForge
For those who are unfamiliar with the installation / newbie, it is highly recommended to install stability matrix as the platform to install various webui (unfortunately reforge not available): https://github.com/LykosAI/StabilityMatrix
Highly recommended to use Forge webui as alternative, automatic a1111 webui is also good to use but it will be slower than Forge. Other image generation tools should be able to use the lora too as I believe it is universal, but I can't verify as I do not use those.
Checkpoint selection
My lora / dora is trained exclusively on specific Pony model. Hence, if wanted to generate image with high resemblance, it is highly recommended to use the said model.
The latest trained lora will be using only_porn (Pony version)
Some additional notes:
a) For 3010xx-nc model, use the version 2.6. Version 3.0 and above will result is losing of resemblance as i noticed it lean more towards western face.
b) For only_porn, ensure to download the Pony version instead of SDXL version.
c) Duchaiten trained lora might also loses some resemblance when using the latest checkpoint version as I trained it on previous version.
Webui extension
a) adetailer (COMPULSORY): https://github.com/Bing-su/adetailer
usable for reforge, forge, automatic a1111
compulsory to achieve the resemblance (face_yolov8n.pt), increase detection to 0.7 & increase denoising to 0.4.
additionally, it is highly recommended to install the eye adetailer to improve further on resemblance: https://civitai.com/models/334668/eye-detailersegmentation-adetailerreduce detection to 0.3, denoising strength below 0.4 to prevent bad generation. (personally I am using 0.35).
b) if you are using reforge, it is also recommended to enable automatic cfg.
Recommended Lora
From my testing, my lora is compatible with most of other lora available in civitai.
To enhanced on the image quality & realism, following lora is recommended:
a) lighting slider: https://civitai.com/models/502254/lighting-slider
weight: -0.5 until -1.5, my standard usage is -0.75.
use to reduce the brightness and saturation which is common with alot of pony model.
b) contrast slider: https://civitai.com/models/536749/pony-contrast-adjuster
weight: -0.5 until -0.75, my standard usage is -0.75.
use in conjunction with lighting slider for the same purpose to improve realism.
c) detail slider (optional): https://civitai.com/models/402462/detail-slider-lora-or-ponyxl-sdxl
can improve detail in image without affecting the character, if affected character, lower the weight or remove this lora.
d) background detailer (optional): https://civitai.com/models/422243/backgroundscenery-detailer-pony
enhance background for better image.
e) pony good hand (optional): https://civitai.com/models/450321/good-hands-for-pony
improve hand gesture & prevent some horrific hand abomination which i noticed might occur with some of my lora in low chance. However, I do noticed tendency of hand touching face if you do not specify further action.
f) zPDXL & z-PDXL PG rating: https://civitai.com/models/332646/pony-pdxl-negative-embeddings
these embeddings can create better image and allow easier pg rating image to be created.
Prompting
example:
score_9,score_8_up,score_7_up,score_6_u,score_5_up,
zPDXL3,zPDXLpg,(photo /medium/:1.2),best quality,perfect quality,sharpfocus,cinematic,masterpiece,epic,gorgeous,very aesthetic,intricate detail,8k,highres,
censored,
<lora:lighting Slider_alpha1.0_rank4_noxattn_last:-0.75><lora:pony-contrast+saturation_adjust-mki:-0.75>realistic lighting,dynamic lighting,<lora:BackgroundDetailerV3:0.9><lora:StS_PonyXL_Detail_Slider_v1.4_iteration_3:0.8>
adult female,1girl,realistic skin,detailed skin,highres skin texture,perfect anatomy,<lora:pony_good_hands:0.5>good_hands,perfect hand,female hand,perfect face,(symmetrical eyes),(perfect eyes),(perfect iris),realistic iris,small iris,symmetrical iris,solo,alone,
[Insert my lora here & trigger prompt]
[other prompt & lora to generate the image you want]to enhance similarity, it is better to specify races, hairstyle, color of the eye, body shape (breasts size, etc).
from my own testing, my lora / dora is very compatible with other lora available as long as the other lora is not overtrained on face (which can ruin the compatibility) or badly trained lora (heavily affect the image ignoring prompting).
Troubleshooting
Sometime, you might get broken image when generating.
(Example: image being too bright / broken brightness. or simply black image with red and dark spot, purely noise)
Usually these error will occur when using in combine with other lora / dora or too much prompting, to fix the error, you may attempt to lower the weight of my lora / dora or changing the arrangement of prompt / cleaning up prompt. If no matter what you tried and the error still remain, it is likely the other lora is badly trained in compatibility.
Conclusion
As there are many users using different setup, I might not foresee any issue occuring with my lora, if you do use my lora and do not achieved intended result, do let me know and I see if I am able to troubleshoot the issue.
Thank you for all support on my lora / dora.