<h1 id="introduction-to-beholder-vision-s4t3rq2cl">Introduction to Beholder Vision</h1>This is an initial test LoRA for Pony, derived from my <a target="_blank" rel="ugc" href="https://civarchive.com/models/989088/l3n4s-beholder-vision">Beholder Vision LyCORIS-LoCon</a> for <a target="_blank" rel="ugc" href="https://civarchive.com/models/795765?modelVersionId=889818">Illustrious-XL v0.1</a>, trained using the foundational parameters outlined in my <a target="_blank" rel="ugc" href="https://civarchive.com/articles/9376/l3n4s-crash-course-in-on-site-lora-training">Crash Course in LoRA Training (On-Site Edition)</a>.Nov 28, 2024 Beholder Vision began as an experimental LyCORIS-LoCon, trained on 126 meticulously curated images of aesthetic anime art.The concept: create “eye-candy” results, with flexibility to adapt to a variety of creative needs.<h2 id="usage-recommendations-zgvbv5f43">Usage Recommendations</h2>Pony Ver. Parameters (Recommended as of Dec. 6, 2024):<ul><li>Strength: Set between 1.0–1.2 (based on extensive testing, I strongly recommend 1.05).</li><li>Sampling Method: DPM++ 2M SDE (Karras)</li><li>Steps: 25–35 steps.</li><li>CFG: 7.</li></ul><h2 id="changelog-and-updates-qquyudtek">Changelog and Updates</h2>Dec 6, 2024 Version <a target="_blank" rel="ugc" href="https://civarchive.com/models/989088?modelVersionId=1133588">3o</a> is here!<ul><li>Full support for v-prediction models such as <a target="_blank" rel="ugc" href="https://civarchive.com/models/833294?modelVersionId=1093948">NoobAI-XL V-Pred-0.65S-Version</a>, which this version is based on.</li></ul>Dec 4, 2024 Version 2.1o is now available! This marks a complete rework of the initial test version for Pony. It has been trained for 5,490 steps, making it competitive with the original Beholder Vision.<ul><li>Despite its extensive improvements, the new LoRA is just 54.8MB—half the size of previous Pony iterations of the model and vastly enhanced in every aspect.</li></ul>Dec 2, 2024 The first version of Beholder Vision for FLUX.1 [dev], 1.3o, has officially released!Nov 30, 2024 I’m excited to announce Beholder Vision <a target="_blank" rel="ugc" href="https://civarchive.com/models/989088?modelVersionId=1114741">1o</a> (and its successor, <a target="_blank" rel="ugc" href="https://civarchive.com/models/989088?modelVersionId=1115716">1.1o</a>)!<ul><li>1o debuts as an experimental GLoRA, while 1.1o returns to classic LoCon, delivering improved performance at a smaller file size.</li><li>Now natively compatible with other SDXL models, including <a target="_blank" rel="ugc" href="https://civarchive.com/models/833294?modelVersionId=1022833">NoobAI-XL Epsilon-pred 1.0-Version</a>.</li></ul>This version represents a significant leap forward over the <a target="_blank" rel="ugc" href="https://civarchive.com/models/989088?modelVersionId=1108064">0.9beta</a>, thanks to an upgraded training dataset and finely tuned parameters. (Details on the dataset and parameters will be shared if there’s interest.)

【LoRA / FLUX.1 & Pony】L3n4's Beholder Vision

Using a smaller portion of my initial dataset, I managed to make a LoRA for Beholder Vision targeted at Flux.1 [dev].

beholder_vision_1.3o.safetensors

1.3o FLUX.1

2.1o P-XL

1.4o P-XL

1.2o P-XL

Introduction to Beholder VisionThis is an initial test LoRA for Pony, derived from my <a href="https://civitai.com/models/989088/l3n4s-beholder-vision" rel="ugc noopener noreferrer nofollow" target="_blank">Beholder Vision LyCORIS-LoCon</a>, trained using the foundational parameters outlined in my <a href="https://civitai.com/articles/9376/l3n4s-crash-course-in-on-site-lora-training" rel="ugc noopener noreferrer nofollow" target="_blank">Crash Course to On-Site Training</a>.Nov 28, 2024Beholder Vision began as an experimental LyCORIS-LoCon, trained on 126 meticulously curated images of aesthetic anime art.The concept: create “eye-candy” results, with flexibility to adapt to a variety of creative needs.Usage RecommendationsPony Ver. Parameters (Recommended as of Dec. 6, 2024):<ul><li>Strength: Set between 1.0–1.2 (based on extensive testing, I strongly recommend 1.05).</li><li>Sampling Method: DPM++ 2M SDE (Karras)</li><li>Steps: 25–35 steps.</li><li>CFG: 7.</li></ul>Changelog and UpdatesDec 6, 2024Version <a href="https://civitai.com/models/989088?modelVersionId=1133588" rel="ugc noopener noreferrer nofollow" target="_blank">3o</a> is here!<ul><li>Full support for v-prediction models such as <a href="https://civitai.com/models/833294?modelVersionId=1093948" rel="ugc noopener noreferrer nofollow" target="_blank">NoobAI-XL V-Pred-0.65S-Version</a>, which this version was trained on.</li></ul>Dec 4, 2024Version 2.1o is now available! This marks a complete rework of the initial test version for Pony. It has been trained 5,490 steps, making it competitive with the original Beholder Vision.<ul><li>Despite its extensive improvements, the new LoRA is just 54.8MB—half the size of previous Pony iterations of the model and vastly enhanced in every aspect.</li></ul>Dec 2, 2024The first version of Beholder Vision for FLUX.1 [dev], 1.3o, has officially released!Nov 30, 2024I’m excited to announce Beholder Vision <a href="https://civitai.com/models/989088?modelVersionId=1114741" rel="ugc noopener noreferrer nofollow" target="_blank">1o</a> (and its successor, <a href="https://civitai.com/models/989088?modelVersionId=1115716" rel="ugc noopener noreferrer nofollow" target="_blank">1.1o</a>)!<ul><li>1o debuts as an experimental GLoRA, while 1.1o returns to classic LoCon, delivering improved performance at a smaller file size.</li><li>Now natively compatible with other SDXL models, including <a href="https://civitai.com/models/833294?modelVersionId=1022833" rel="ugc noopener noreferrer nofollow" target="_blank">NoobAI-XL Epsilon-pred 1.0-Version</a>.</li></ul>This version represents a significant leap forward over the <a href="https://civitai.com/models/989088?modelVersionId=1108064" rel="ugc noopener noreferrer nofollow" target="_blank">0.9beta</a>, thanks to an upgraded training dataset and finely tuned parameters. (Details on the dataset and parameters will be shared if there’s interest.)

Files with SHA256: b51b0b13ae1d28df037770320103fda45d2a33edc08a9ff7ca37c7aa2aea91a4