<p>Hey everyone! 👋</p><p>Presenting my <strong>Annie Hughes LoRA from <em>The Iron Giant</em></strong>.</p><p>____________________________________________________________________________________________</p><h1 id="moss-update">MOSS-Update</h1><p>This is my second attempt at training an Annie MOSS LoRA. This time, I let Civitai's autotagger and trainer handle most of the work.</p><p>The results are decent, although the automatic captioning wasn't always the best. So, if you get some unexpected or unusual results, you might need to adjust the prompts a little.</p><p>Just like the Anima version, this LoRA supports both of Annie's familiar looks:</p><ol><li><p><strong>Waitress Look</strong></p></li><li><p><strong>Home Look</strong></p></li></ol><p>I'd really appreciate it if you shared your generations and left a review. Feedback would be very helpful! ❤️</p><h2 id="trigger-word">Trigger Word</h2><p><code>annieh</code></p><h2 id="appearance">Appearance</h2><pre><code>a woman with wavy brown hair in an updo, green eyes</code></pre><h2 id="default-outfits">Default Outfits</h2><p><strong>Waitress Look</strong></p><pre><code>She is wearing a pink waitress uniform, a white apron that ties at the waist, and white loafers.</code></pre><p><strong>Home Look</strong></p><pre><code>She is wearing a blue long-sleeve shirt with black leggings, a purple apron, and black loafers.</code></pre><p>As usual, you don't need to prompt for footwear unless you're going for full-body generations.</p><h2 id="my-(and-suggested)-generation-settings">My (and Suggested) Generation Settings</h2><p>I tried a couple of combinations, but Euler didn't give me the best results. Feel free to experiment with different sampler and scheduler combinations.</p><ul><li><p><strong>Scheduler:</strong> er_sde</p></li><li><p><strong>Sampler:</strong> beta</p></li><li><p><strong>CFG:</strong> 3.0</p></li><li><p><strong>Steps:</strong> 50</p></li></ul><h2 id="model-compatibility">Model Compatibility</h2><p>The LoRA was trained on <strong>MOSS-Image-Base</strong>, but all my example generations were made using the <strong>MOSS-Image-SFT INT8 Pruned</strong> model.</p><p>The LoRA was trained on <strong>Anima 1.0</strong> and works particularly well with two of Annie's default looks from the movie:</p><ul><li><p><strong>Waitress look</strong></p></li><li><p><strong>Home look</strong></p></li></ul><p>The LoRA is trained primarily for Annie's <strong>character identity, appearance, and these two outfits</strong>, while still allowing you to use different outfits and poses.</p><p>_____________________________________________________________________________________________</p><h2 id="base-model">Base Model</h2><p><strong>Anima 1.0 - Base checkpoint</strong><br /><code>anima-base-v1.0</code></p><p>The LoRA should also work well with other <strong>Anima-based checkpoints</strong>.</p><h2 id="recommended-weight">Recommended Weight</h2><p><strong>0.9</strong></p><h2 id="trigger-word">Trigger Word</h2><p><code>anniehij</code></p><h2 id="style">Style</h2><pre><code>2d animation, flat colors</code></pre><hr /><h2 id="appearance-waitress">Appearance - Waitress</h2><pre><code>auburn hair, updo, green eyes</code></pre><h3 id="default-outfit-waitress">Default Outfit - Waitress</h3><pre><code>pink waitress uniform, knee length, white collar, white waist apron, short sleeves</code></pre><h3 id="optional-full-body-poses">Optional - Full Body Poses</h3><pre><code>white footwear</code></pre><hr /><h2 id="appearance-home">Appearance - Home</h2><pre><code>auburn hair, green eyes, hairband</code></pre><h3 id="default-outfit-home">Default Outfit - Home</h3><pre><code>blue long sleeve shirt, dark purple apron, black capri leggings</code></pre><h3 id="optional-full-body-poses">Optional - Full Body Poses</h3><pre><code>black footwear</code></pre><hr /><h2 id="using-the-tags">Using the Tags</h2><p>The <strong>appearance</strong> and <strong>default outfit</strong> tags can be used separately depending on what you want to generate.</p><p>For example, you can use the character's appearance tags while changing the outfit completely, or combine the appearance with either of the default outfits.</p><p>The LoRA should also work with <strong>different outfits and poses while retaining Annie's identity</strong>.</p><p>For the full-body examples, the footwear tags can be added when needed.</p><h2 id="suggested-generation-settings">Suggested Generation Settings</h2><p>My testing was done using a <strong>Turbo LoRA</strong> with the following settings:</p><ul><li><p><strong>LoRA Weight:</strong> 0.9</p></li><li><p><strong>Sampler:</strong> Euler a</p></li><li><p><strong>Scheduler:</strong> Simple</p></li><li><p><strong>Steps:</strong> 12</p></li><li><p><strong>CFG:</strong> 2.5</p></li></ul><p>These settings worked well for my testing, but other Anima-compatible settings and checkpoints may require some adjustment.</p><h2 id="training-examples">Training / Examples</h2><p>I tried to tag the training images with <strong>accurate metadata</strong>, and the example images were generated using recorded seeds to make the results easier to reproduce.</p><p>You can refer to the example images for additional generations, outfits, poses, and variations.</p><p>Hope you enjoy the LoRA! ❤️</p>

Files with SHA256: d036659abe52aa042f650f6efbfd5b8a88ca82a1b5bdb08c74ad89feb4ffc5cf