<h2 id="asagiri-shiori-from-tamako-market"><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/6d25bb1c-6107-4124-a701-6de558898af1/original=true/6d25bb1c-6107-4124-a701-6de558898af1.jpeg" />💥 Asagiri Shiori from Tamako Market</h2>— She is a quiet and reserved girl who has a hard time expressing herself, even when she really wants to say something. She can seem distant at first, but she is actually very polite and kind, just extremely shy and awkward when talking to other people. That awkward side of her is probably one of the most memorable parts of her character, especially once she slowly starts opening up. A simple but really cute character :)<h2 id="greatertamako-market-collectionless"><a target="_blank" rel="ugc" href="https://civarchive.com/collections/17567526">&gt;Tamako Market COLLECTION&lt;</a></h2><h1 id="turn-off-your-rxxxx-filter-to-see-hidden-juicy-loras-in-greaterhereless">⚠️Turn off your R/X/XXX Filter to see hidden JUICY Loras in <a target="_blank" rel="ugc" href="https://civarchive.com/collections/13797500">&gt;HERE&lt;</a></h1><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/e4c11a1c-1328-4124-9572-eee4f03663f0/original=true/e4c11a1c-1328-4124-9572-eee4f03663f0.jpeg" /><h2 id="how-i-generate-my-images">🖼️ How I Generate My Images</h2>I made a full article that explains how I generate my images — including my settings, prompting tips, and even my potato PC setup 😅👉 <a target="_blank" rel="ugc" href="https://civarchive.com/articles/17562">Check it out here</a> if you’re curious or want to try the same method.<h2 id="nsfw-gallery-(18+)">🔞 NSFW gallery (18+)</h2>Check my account <a target="_blank" rel="ugc" href="https://sociabuzz.com/zetsubou">here </a>to see the spicy showcase that I can't show here, if any.<h2 id="prompt-tips">🎨 Prompt Tips</h2><h3 id="getting-the-correct-hair-color">🌑 Getting the correct hair color</h3>Even though her dataset was trained using <code>grey_hair</code>, prompting it tends to make her hair noticeably lighter than her canon color.For better results, try using: ✅ Positive prompt: <code>dark_grey_hair</code>For some reason, this gives me a color much closer to her actual appearance than <code>grey_hair</code>.<h3 id="extra-pocky-hair-ornaments">🍫 Extra Pocky hair ornaments</h3>Her hair ornament may occasionally generate more than two Pocky-like pieces. There is not much special prompting needed for this one. In most cases, a small retry is enough to get the correct amount.<h3 id="reducing-wall-like-details-in-outdoor-scenes">🌳 Reducing wall-like details in outdoor scenes</h3>The raw dataset has very little nature scenery and mostly consists of indoor locations or backgrounds with buildings. Because of this, outdoor prompts can occasionally inherit wall-like structures even when the scene is supposed to be somewhere like a park or other natural environment.If this happens, try using: 🚫 Negative prompt: <code>wall, indoors</code>This will not completely remove the dataset bias, but it can help reduce those unwanted structures.<h3 id="preventing-multiple-school-bags">🎒 Preventing multiple school bags</h3>Prompting her school bag can sometimes result in more than one bag appearing.I recommend keeping the positive prompt simple: ✅ Positive prompt:<code>school_bag</code> 🚫 Negative prompt: <code>multiple_bags</code>Using only <code>school_bag</code> instead of stacking several bag-related tags seems to work better for me.<h3 id="cardigan-and-default-outfit-bias">🧥 Cardigan and default outfit bias</h3>Almost all of her source material shows her wearing the open black cardigan, so it can occasionally appear even when you do not want it. This version was trained with cardigan-free uniform generation in mind, so the uniform itself can still work without it.If the cardigan keeps appearing, try putting the unwanted cardigan-related tag in your negative prompt before lowering the LoRA weight.<h3 id="suppressing-baked-in-pleated-skirt-details">👗 Suppressing baked-in pleated skirt details</h3>Her pleated skirt seems to be learned a little too strongly and can occasionally influence other outfits.If pleated details keep appearing on clothes that should not have them, try using: 🚫 Negative prompt: <code>pleated_skirt</code>You can also gradually lower the LoRA weight if the detail is still persistent. I recommend trying the negative prompt first, since lowering the LoRA too much can also weaken her face, hair, and overall character identity.<h3 id="preventing-the-school-emblem-from-leaking">🐰 Preventing the school emblem from leaking</h3>Her uniform has a small bunny-like school logo, but the source coverage for it is not good enough for reliable fine details. This version instead focuses more on preventing the emblem from appearing automatically on unrelated outfits.If it still appears on another outfit, try using: 🚫 Negative prompt: <code>school_emblem</code>The emblem itself may also look inconsistent when you intentionally prompt it, simply because it is a very small feature with limited dataset coverage.<h3 id="lora-strength-and-outfit-flexibility">⚖️ LoRA strength and outfit flexibility</h3>The raw dataset almost exclusively shows her wearing the same school uniform with an open black cardigan, so completely different outfits may still need stronger prompting.A weight of <code>1</code> can still work with different outfits, especially when the new outfit is clearly defined.Only lower the LoRA weight gradually when default outfit details keep leaking even after using clear outfit tags and negative prompts. Lowering it too much may also weaken her face, hair, and overall character identity.<h2 id="lora-project">📊 Lora Project</h2>If you want to know what characters that I'm covering next you can check these link : 👉 <a target="_blank" rel="ugc" href="https://docs.google.com/spreadsheets/d/1oM6Xk0k5sBCw2CQfIFMzQQuuC6lHdOOnw3zES9phBm8/edit?gid=1864372839#gid=1864372839">Personal Project</a> 👉 <a target="_blank" rel="ugc" href="https://docs.google.com/spreadsheets/d/1k15ya9i0TjAHJnBpVfPSsjHsngpjkLsAZ11IcOT0hc0/edit?gid=0#gid=0">Commision Project</a>If this model sparked something in you, show some love with a like, review, or buzz. Now go— Be wild

Asagiri Shiori | Tamako Market/たまこまーけっと

🎬 Anime and movie screenshots with official art support<pre><code>This LoRA was trained using a mix of anime screenshots, movie screenshots, and Megami Magazine scans. The movie screenshots received some color correction to keep them closer to the rest of the dataset, while some magazine images were adjusted to reduce unnecessary clutter around the character.</code></pre>🍫 Adjusted hair ornament colors<pre><code>Her Pocky-like hair ornament appears with different colors in the source material. I adjusted them so the dataset consistently uses the chocolate-colored version, giving the trainer a cleaner and more consistent signal for such a small accessory.</code></pre>🏫 Limited environment variety<pre><code>The raw dataset mostly consists of indoor scenes or backgrounds containing buildings and walls, with very little proper nature scenery. Because of this, some environment bias is still expected, especially when generating outdoor scenes.</code></pre>🧥 Main outfit bias with cardigan-free support<pre><code>Almost all of the raw dataset shows her wearing her main school uniform together with the open black cardigan, so some bias toward it is unavoidable.

Even with that limitation, this version was specifically optimized to still be able to generate her school uniform without the cardigan.</code></pre>🐰 School emblem prioritizes flexibility<pre><code>Her uniform has a small bunny-like logo, but the feature has poor coverage and is too small to reproduce consistently.

Instead of focusing too much on accurately reproducing such a tiny detail, I prioritized preventing the emblem from becoming heavily baked into the character and appearing on every other outfit.</code></pre>🎨 Hair color behavior<pre><code>Her hair color was tagged using grey_hair throughout the dataset.

Weirdly enough, prompting grey_hair does not reproduce her canon hair color very well and usually makes it noticeably lighter. In my testing, dark_grey_hair works much better during generation.</code></pre>As usual, the final result can still depend on your checkpoint, prompt, LoRA strength, CFG scale, camera angle, and generation settings.

Files with SHA256: cd0ec658a9d64cf27a1fd941f5602530dd42ecb7d503b710cc843df36c3101cb