Loading
Have a ComfyUI workflow? Get a readable Workflow Doc in your browser.Make a workflow readable.Try it now ↗
Krea 2 facial-expression control LoRA trained on adult East Asian female portrait references. No trigger word is required. Describe expressions naturally using gaze direction, eyelid openness, eyebrow position, mouth...
Model source
Source description
This version is a major refinement over the original Nuance Female Emotion LoRA.
What’s Improved
Stronger facial mechanics control V3.1 focuses much more on precise, readable facial actions instead of broad emotion labels. You’ll get clearer results on:
Eye direction and eyelid position
Brow tension and asymmetry
Mouth shape (parted, pressed, asymmetrical smiles)
Tongue position and subtle lip gestures
Micro-expressions and held emotions
Sources
1 sourceModel artifacts
1 artifactkrea_2_nuance_female_emotion_epoch_10.safetensors
safetensors · BF16 · 218 MB · SHA-256 f6efa519820a…b6a7 · Civitai
DownloadSource excerpts
1 excerptSource context: 389 downloads · Type LORA · Base model Krea 2
Better expression range Compared to V1, this version handles a wider spread of nuanced states more reliably:
Sleepy / half-lidded / just-awake
Coy and over-the-shoulder glances
Uneven winks
Restrained laughter
Soft relief and quiet vulnerability
Micro disgust / aversion
Soft cognition / realization
Puffed cheeks and playful asymmetry
Cleaner trigger behavior The new trigger nuexpress is more consistent. Pairing it with specific facial descriptions (the way the training captions were written) activates the LoRA more reliably than the broader prompting style needed for V1.
Higher data quality
171 carefully filtered pairs
No exact duplicates
Controlled intensity ladders
Stronger focus on East Asian facial features and natural skin texture
Better balance between subtle and exaggerated expressions
Recommended Use
For best results, start prompts with:
nuexpress, early-20s East Asian woman, [specific facial mechanics]...
Describing the actual eye, brow, and mouth positions works better than just writing “sad” or “happy.”
