It's my 1st LoRa, so it may be not that good. Feedback appreciated 🫡<h2 id="character-hu38at6es">Character</h2>Liza from Death March to the Parallel World Rhapsody.Age: 18Type: Lizard GirlSignature Moves:<ul><li>frowning, looking worried</li><li>standing in the background</li></ul><h2 id="recommended-usage-ukh9oso9i">Recommended usage</h2>weight:<ul><li>1.0 - to reproduce all details of her dresses</li><li>0.8-1.0 - for more flexibility</li></ul>Dresses and positive prompt<pre><code># 1. dress: "crop top + miniskirt"
catDMLiza, grey crop top, stomach, grey miniskirt, metal collar, collar, red scales on neck, red scales on shoulders, red scales on arms, red scales on legs, lizard tail

# 2. dress: uniform, brown jacket
catDMLiza, brown jacket, red aiguillette, long sleeves, black pants, black gloves, black belt, red scales on neck, lizard tail

# Details (add if you like)
# 1. collar
metal collar, collar

# 2. holding spear
weapon, spear, holding, holding weapon
# or just
spear, holding weapon

# 3. scales
red scales on shoulders, red scales on neck, red scales on arm, red scales on legs

# 4. character details that are pruned into catDMLiza (add if something gets lost)
small breasts, blue eyes, hair between eyes, red hair, very long hair, low-tied long hair, hair bow</code></pre>Negative prompt<pre><code># reduce anime coloring
anime coloring, anime screencap

# reduce blurriness
blurry, blurry background

# misc, (covered spear is shown on some training images)
dark, long neck, covered spear

# all
blurry background, blurry, anime coloring, anime screencap, dark, long neck</code></pre><h3 id="showcase-images-xk5umpw9g">Showcase-Images</h3>created with:<ul><li>A1111 ReForge, finesse v3, 28-35 steps, HiResFix x 1.5, 20 steps, 0.25 denoise</li><li>ComfyUI-Workflow: <a target="_blank" rel="ugc" href="https://civarchive.com/models/741620?modelVersionId=1240954">https://civarchive.com/models/741620?modelVersionId=1240954</a>, scheduler=sgm_uniform</li></ul><h3 id="funny-notes-re934ndmr">Funny notes</h3><ul><li>the LoRA is a bit overtrained, but can reproduce most details quite consistently and correctly</li><li>the collar might be showing itself more if the weight of tags go up</li><li>the red scales of Liza are not consistently placed across all images of the dataset, so it might differ on arms and legs. In some pictures there were no scales at all on the arms and legs. wtf</li></ul>

Files with SHA256: ef08d0825b216965c9d7a8cccc9b0e32e39bd2a158afcd21dde3d5cbb5b97e12