Please consider joining my Patreon so I can keep most of my work available for everyone - I'll be releasing early access models there for a cheaper monthly fee than buying them here individually, and also will be providing exclusive models and more!

For business inquiries, commercial licensing, custom models, and consultations, please get in touch at hello@zavyai.com or sales@zavyai.com. You can also contact me here through CivitAI DM or join my Discord.

Consider using Euler A for base image generation (including hires fix), and upscale further with DPM++ 3M SDE Exponential or DPM++ 2M Karras

For anyone wondering - I will not be supporting the merging of LoRAs such as Lightning, which quickens rendering speed at a loss of significant quality and flexibility, into checkpoints. Read my further thoughts in this article.

As a reminder, due to the various workflows, internal settings, UIs, etcetera... copying a prompt will not necessarily get you the same result as another.

Introduction

Introducing a new model iteration based on Pony Diffusion V6 XL, tailor-made for western art enthusiasts, especially those in animation and game design. Unlike its predecessor, which often fell short in rendering backgrounds with richness and detail, this version rectifies that flaw nearly entirely. While still capable of emulating the diverse styles encompassed by PonyXL, its sweet spot lies in capturing the essence of western art styles.

Moreover, this iteration accommodates SDXL LoRAs much better than PonyXL does, allowing users to extend their range of LoRAs used. Though not a fail-safe guarantee of working as intended, you will tend to keep good image results, unlike the hit-or-miss outcomes of PonyXL. It is advisable to elevate the weight of SDXL LoRAs, from the usually recommended 0.6 to a full 1.0.

Continue reading for more information and some tips and tricks.

I kindly request that you share your creations both here and on my Discord server, as I would greatly appreciate the opportunity to see them and motivate me to spend more time in further ventures here.

Pros and cons

Pros

• Perfected for more western art, game and animation art. Think Dungeons & Dragons, World of Warcraft, Disney, Avatar (series), etcetera.

• Much better saturation than base model.

• Much better backgrounds, more contrast and detail.

• Solely landscape pictures actually start looking good, compared to base.

• Just as capable as Pony with regards to ... well you know.

Cons

• Leans heavily towards female subjects.

• You can get somewhat close to photorealism, but not recommended, use ZavyChromaXL for that.

Roadmap

1. Training ZavyChromaXL and ZavyFantasiaXL continuously.

2. Pioneering uncharted LoRA subjects (withholding specifics to prevent preemption).

Tips

• Make sure you add score_9, score_8_up, score_7_up at the front of your prompt. While not as required as base PonyXL, it still helps add more quality to the image.

• Don't use the SDXL refiner.

• SDXL LoRAs are not always providing the expected result, it needs higher weight to see results than they do on SDXL, and some SDXL LoRAs will not work regardless. It will work much better than base PonyXL does though. Make sure that you use strong danbooru tags sparingly, or you will easily outweigh most SDXL LoRAs.

• ComfyUI is the UI I use for my SDXL images.

• SD1.5 LoRAs won't work in SDXL.

• Make sure to use danbooru tags alongside the normal SDXL prompt.

• To maintain optimal results and avoid excessive duplication of subjects, limit the generated image size to a maximum of 1024x1024 pixels or 640x1536 (or vice versa). If you require higher resolutions, it is recommended to utilise the Hires fix or similar.

• To better understand the preferences of the model, individuals are encouraged to utilise the provided prompts as a foundation and then customise, modify, or expand upon them according to their desired objectives.

Prompts

Recommended positive prompt: score_9, score_8_up, score_7_up, sentence(s) about what you want, danbooru tags,

Recommended negative prompts: source_pony, source_furry, malformed, deformed, worst quality, simple background, bad hands, bad feet, watermark, censored,

Exchange source_pony, source_furry for precisely what you do not want to see from the Pony model.

For danbooru tags, see:

danbooru

Recommended settings

• sdxl_vae.safetensors (baked in).

• Euler A for base gen, DPM++ 3M SDE Exponential or DPM++ 2M Karras for upscale (USDU especially).

• Steps 30~50 (lower range for DPM, higher range for Euler).

• Hires upscaler: UltraMix_Balanced or Remacri.

• Hires upscale: Whatever maximum your GPU is capable of, but preferably between 1.5x~2x.

• CFG scale 4-10 (preferably somewhere around cfg 6-7)

Lightning LoRA specific settings:

• Not yet tested.

Model recipe

Trained from Pony Diffusion V6 XL.

Social media

You are welcome to join my recently created Discord server, where we can engage in discussions, share our experiences in AI, and showcase the things we’ve made with AI. You are encouraged to join and ask any questions or seek additional tips and tricks related to my models or AI in general. Your participation would be greatly appreciated.