<h3 id="introducing-v3.0a:-the-ultimate-refinement-of-the-line-art-style!-4kp9dr39a">Introducing V3.0A: The Ultimate Refinement of the Line Art Style! 🎨</h3>Hello everyone!Following the foundational V1.5 and the more developed V2.0, we are thrilled to announce the definitive and final version of this series: V3.0A.This version represents the culmination of our work, inheriting all the strengths of its predecessors while resolving their known limitations. It's the most polished and powerful version to date.<h3 id="key-upgrades-in-v3.0a:-hy5ml5p00">Key Upgrades in V3.0A:</h3><ul><li>Enhanced Long-Range Clarity: We have completely fixed the issue from V2.0 where distant scenes and characters would have blurry or chaotic lines. Now, every line, whether in the foreground or background, is rendered with crisp, clean precision.</li><li>Optimized for High Resolution: V3.0A is specifically built for generating at 1024px and higher. It maintains exceptional sharpness and line smoothness, fully resolving the minor flatness issues that could appear in V2.5 at this resolution.</li><li>Flexible Color Control: The model perfectly inherits the color capabilities. It will generate in color if specified in your prompt. To achieve the signature, striking black-and-white comic style, simply add the <code>monochrome</code> , <code>grayscale</code> keywords to your prompt.</li><li>Superior Compatibility: This model has a more stable and refined style, making it highly compatible when mixed with other LoRAs. This greatly reduces the chance of style clashing, giving you more freedom to combine and create.</li></ul><h3 id="recommended-settings-(usage-guide):-4rq1ifj1x">Recommended Settings (Usage Guide):</h3><ul><li>Trigger Words: <code>V3.0A</code> , <code>Monochrome</code> , <code>grayscale</code></li><li>Recommended Weight: <code>0.75 - 0.9</code> .</li><li>CFG Scale: <code>4 - 8</code> . The optimal value is around <code>6</code> .</li><li>Sampler:<ul><li><code>DPM++ 2M</code> : Recommended for sharper lines, ideal for character portraits and close-up shots.</li><li><code>Euler</code> : Recommended for more dynamic and varied lines, perfect for distant views and creating expressive landscapes or backgrounds.</li></ul></li></ul><hr /><h3 id="a-special-thank-you-u6qb7n5n0">A Special Thank You</h3>We would like to extend our heartfelt gratitude to everyone who participated in the previous bounty. Your ideas and prompts were invaluable and directly contributed to the style and direction of V3.0A.Thank you, and have fun creating!————————————————————————————————————————First of all, I would like to thank the master Tentacles_Riders for the inspiration provided by the pictures he made. I have to say that the aesthetic is really great.————————————————————————————————————————None of this would have been possible without the incredible support of <a target="_blank" rel="ugc" href="https://civarchive.com/user/Tentacles_Riders">Master Tentacles_Riders</a>, who generously provided 100,000 buzz. His contribution was the foundation that ensured my LoRA could be successfully produced. Thank you for your invaluable help—it means the world to me! 🌟————————————————————————————————————————<h3 id="key-upgrades-in-version-2.0:-2fmnvbwp4">Key Upgrades in Version 2.0:</h3>Running parameters: cfg4-6 thin lines, cfg7-10 thick lines, if you want to adjust, you can add corresponding line prompts, step number 24-40, recommended 36, Sampler recommends DPM++2M, if it is not flat enough, please adjust the prompts, add, flat painting, celluloid, etc. Lora still has a defect that the faces of people in the long-range view are blurred, the lines are chaotic, and the closer they are, the clearer they are. Some long-range pictures need hi-fix twice.Prompt word V2.0 can keep the original color, and the monochrome will become black and white. It has not changed because the color is written in the prompt word, or it can (monochrome)<ul><li>Optimized Operational Experience: Addressed multiple known issues for smoother and more stable performance.</li><li>Enhanced Visuals: Markedly improved image quality and overall visual presentation for a superior look and feel.</li><li>Expanded Creative Freedom: Introduced support for characters of various sizes and perspectives, enabling richer creative expression.</li><li>Improved Detail Handling: Enhanced line fineness and detail rendering for more sophisticated-looking creations.</li></ul>————————————————————————————————————————<h3 id="ntroducing-v1.5:-enhanced-stability-refined-lines-and-a-distinctive-comic-style!-0c2klkxwn">ntroducing V1.5: Enhanced Stability, Refined Lines, and a Distinctive Comic Style! 🎨</h3>Hello everyone! I'm thrilled to announce an update to our LoRA model. Responding to a few instabilities and the occasional uneven line thickness in V1.0, and after some intensive work with more focused training content, I've developed V1.5!What's New and Improved in V1.5?<ul><li>Enhanced Stability &amp; Consistency: We've addressed the previous version's unpredictability. V1.5 offers a more reliable and consistent output.</li><li>Refined Line Work: One of the core focuses was to achieve more uniform and intentional lines.</li><li>Streamlined Training for a Fixed Style: To achieve a distinct and consistent aesthetic, I first generated 2000 images using AI. From this extensive output, I meticulously selected 242 of the best high-quality images. This curated selection then underwent three full training rounds, solidifying its unique artistic signature.</li><li>Distinctive Style Characteristics: V1.5 is defined by:<ul><li>Black and White Stippling/Dot Work : Adding that classic comic texture.</li><li>Clear Comic Influence: Evoking a graphic novel feel.</li><li>Smooth, Flowing Lines: Ensuring a polished look.</li><li>Bold Drawing Style: For impactful visuals.</li><li>Fine Hand-Drawn Lines: Retaining an artistic, crafted quality.</li></ul></li></ul>A Note on Character Depiction &amp; The Road Ahead:This version features 6 distinct characters. While V1.5 has made significant strides in style, I acknowledge there might still be some limitations regarding the breadth of character depiction. Think of this as a beautifully specialized iteration.But don't worry, the journey doesn't stop here! V2.0 is already on the horizon, and it will aim for much broader applicability across a wider range of characters, building on the solid foundation and stylistic strengths of V1.5.————————————————————————————————————————🔥Big launch! The new dot comic style lora is here! Accurately reproduce the classic dot texture, smooth and natural lines, rich light and shadow layers, and easily create an authentic comic atmosphere. Adapted to the IL model, suitable for comic creation, illustration design, avatar production and other scenes. Start your comic creation journey with one click! (Written by ai, I'm really too lazy to write)————————————————————————————————————————This version is available, but it may still be a bit slow. If you want to produce a monochrome embellishment effect in the black and white comic style, please add a prompt word such as "purple hair".A digression, for example: when I was training Sona's picture, I didn't add the "twin tails" label, so if you are not satisfied with the generation, you should use it with the character Lora, or add a reverse prompt wordIf you encounter any problems, please give me feedback, I will integrate and modify it (I feel sorry for my buzz),I wish you a happy use————————————————————————————————————————Key words simplebackground,artist:moriimee,748cmstyle,simplebackground,realistic,halftone,highcontrast,border,(((monochrome)))Fold training parameters: Number of training: 10Number of repetitions: 24Batch size: 4Number of steps: 5760Resolution: 1024LoRA type: loraEnable buckets: yesRandom labels: yesKeep labels: 0Skip segments: 2Flip enhancement: yesUnet LR: 0.00050 (5e-4)Text encoder LR: 0.00005 (5e-5)LR scheduler: cosine_with_restartsLR scheduler period: 1Minimum signal-to-noise ratio Gamma: 5Network dimension (rank): 98Network alpha: 64Noise offset: 0.10Optimizer: AdamW8bitOptimizer parameters: weight_decay=0.1————————————————————————————————————————"This world is a huge alchemy furnace, you and I are all immortals"

Dot Comics

Lora training test<ul><li>It can be used normally</li></ul>First of all, I would like to thank the master Tentacles_Riders for the inspiration provided by the pictures he made. I have to say that the aesthetic is really great.This is a demo version. As a new player, this is my first time training LoRA, and I lack experience. So the generated images are likely not ideal. I didn't fully understand the training process and used the original model of SDXL. I will try my best to train the LoRA style as soon as possible and release the V01 version. I hope everyone can have a great time with it in the future.

Files with SHA256: c756d6c8428cd0635504c73ac16e17f809605347bf7b29771109c4335a5f48af