<a target="_blank" rel="ugc" href="https://www.mage.space/play/c4d2f7081c5145f6827da1783e7c95f2">This model is also available on Mage</a> with free and unlimited generations for October. For more Halloween models, visit: <a target="_blank" rel="ugc" href="https://www.darkinfinity.art/halloween">darkinfinity.art/halloween</a>.I post models on multiple sites, including exclusive ones on specific platforms. Follow me on <a target="_blank" rel="ugc" href="https://x.com/DarkInfinityArt">X (Twitter)</a> or <a target="_blank" rel="ugc" href="https://www.instagram.com/darkinfinityart/">Instagram</a> for new model releases, bounties, and more.<h1 id="the-magic-of-ancien">The Magic of Ancien</h1>Anatomica is a flux LoRA created to replicate the style used for anatomy and medical engravings; however, the style transfers well and works great for any vintage-style image on old parchment.<h3 id="anatomica-v9.0">Anatomica v9.0</h3>I've been experimenting with anatomical since the first LoRA, but for the last three versions (v7, v8, and v9), I've been varying the use of multiple training resolutions. The previous two versions (v7 and v8) weren't different enough from v6 to post; however, the training of version 9.0 seems to have strengthened the LoRA's ability to apply a style without significantly changing the composition of the images. Prompting and trigger words are still the same as described in v6 below, but prompt adherence when using the LoRA is improved. After one final test, I'll share everything I've done and learned about training LoRAs for Flux in an article so others can learn from the Anatomica experiment.<h3 id="anatomica-v6.0">Anatomica v6.0</h3>Version 6.0 has simplified triggers with a much stronger ability to generalize. You need to prompt the trigger word “anatomica” and then one of the three styles outlined below:<ul><li>Use “engraving” to mimic the appearance of an illustration created using an engraving plate. It should have shading and cross-hatching with fine lines work. The style often puts extra details expected of the style, such as plate labels or artist signatures.</li><li>Use “drawing” for simpler lines with less depth and cross-hatching than an engraving.</li><li>Use “illustration” to produce a color illustration with simple lines, similar to an image drawn with dark lines and colored pencils. If it stubbornly sticks to monochrome, you may need to prompt for the color.</li></ul>Example prompts with emphasis on keywords:<ul><li>“An engraving of a child flying a kite on a windy beach, with waves crashing in the background. The kite dances in the sky, tracing loops and spirals. anatomica, &lt;lora:anatomica_v60:1&gt;”</li><li>“A drawing of a group of fishermen casting their lines from a small wooden boat on a calm lake. The early morning mist rises from the water, and birds fly overhead. anatomica, &lt;lora:anatomica_v60:1&gt;,”</li><li>“An illustration of a fox standing still in a snowy field, ears perked, listening for the slightest sound. Snowflakes gently fall, dotting its fur with tiny white specks. anatomica, &lt;lora:anatomica_v60:1&gt;,”</li></ul>The LoRA also transfers other features of anatomical sketches to new subjects. Use phrases that include “labels with lines pointing at … “ to prompt an image with sketched labels. Titles and notes can also be prompted.

Halloween: Anatomica

A flux LoRA created to replicate the style used for anatomy and medical sketches used on engraving plates.Trained using 30 images with minimal tagging in Civitai's LoRA trainer. This version seemed to do best and was pulled off at step 1001. Trained at 0.0005 LR, 10 repeats, epoch 13, 64 alpha/dims.The LoRA responds well when prompted to make an anatomical sketch of something that makes sense in that context but gets confused easily when you prompt totally outside of that domain. Either way, it does some fun images.Future versions should be more flexible and have the ability to generalize so that it can apply the style with a wider variety of prompts.

Files with SHA256: 1be0e92298011b398f752cea1317edc9e9c1d12148b40386ccb7df5da75583b9