My recommended starting point for this series.
Patina E9 is my main pick for adult character work, especially when I want maturity, fatigue, physical presence, and a face that looks like it has lived a little.
These are sibling epochs from the same KREA2 Turbo training line, with overlapping abilities and different tendencies:
Patina E9: start here for stronger character presence, mature faces, visible wear, and heavier moods.
Poise E7: try this when E9 makes the character feel too worn, severe, or old, but you still want substantial maturity and texture.
Muse E5: try this when E7 or E9 repeatedly pushes the character older or more tired than you intended, or when you prefer a softer, more immediately appealing portrait.
These are tendencies, not separate age modes. All three can produce young adults, older women, attractive portraits, and visibly tired characters.
Base model: KREA2 Turbo
What Patina Does Well
Strong mature and older character presence, including ordinary, tired, weathered, and less idealized faces.
Fatigue that can extend into posture and body language rather than appearing only as dark circles.
Good handling of practical clothing, layered outfits, formalwear, accessories, and environmental props.
Clear material differences between fabric, leather, fur, metal, glass, skin, and painted surfaces.
Usable two-person scenes with distinguishable faces and clothing.
Strong full anthropomorphic character rendering, including substantial bodies and practical workwear.
Broad medium responsiveness across watercolor, ink wash, gouache, oil painting, impasto, graphic illustration, and fashion sketch.
E9 has not become locked into realism. It can still produce youthful graphic images, loose ink figures, and flat painted portraits. Its stronger maturity bias depends on the subject and wording.
Known Weaknesses
Age can overshoot. Explicit phrases such as "early thirties" did not consistently prevent older-looking results, particularly with white hair, fatigue, or practical workshop settings.
More character is not always more fidelity. It may add wear, severity, or exhaustion beyond what the prompt asks for.
Gloss is still present. Faces, hands, and exposed skin can retain strong highlights even on the Turbo base.
Beautification has not disappeared. Formalwear, club portraits, and fantasy characters can still drift toward polished faces, styled expressions, or more revealing clothing.
Strict graphic simplification is uneven. Some poster prompts retain dense facial modeling and shiny-looking highlights instead of committing to broad, simple shapes.
Familiar facial archetypes recur within themes. Different outfits do not automatically produce different identities.
Hands and complex contact still need inspection. Detailed fingers and nails do not guarantee correct anatomy.
Two specific failures appeared across all three epochs: a prompt asking for a human woman with subtle tiger traits became a full anthro character, and a crime-boss prompt specifying a bound chest produced exposed cleavage instead. These are real fidelity limitations, even when the resulting images look good.
Best Uses
Mature character portraits, tired or physically worn adults, ordinary older women, grounded workshop scenes, atmospheric paired portraits, full anthro characters, painterly studies, and mature fashion illustration.
Prompting Notes
Be specific about the age, physical condition, clothing, and degree of stylization you actually want. Avoid stacking age and fatigue terms unless you want a stronger effect.
For traditional media, describe the behavior of the medium: transparent washes, ink bleeding, opaque color blocks, or raised palette-knife marks. A successful medium change does not guarantee the same facial identity across images.
Requested sweat, paint sheen, and reserved white paper should be judged differently from unwanted glossy skin.
Tested Settings and Other Bases
The comparison used 66 prompts and 132 images per epoch, with different seeds across epochs. These observations describe the tested samples, not a controlled statistical ranking.
LoRA weight: 1.0
Sampler: Euler
Sampling steps: 8
CFG scale: 1
Resolution: 768 x 1152
Trigger word: none used
Start at 1.0 if you want the behavior described here. Other weights are open to experimentation.
You can also try other KREA2-derived bases. The underlying checkpoint can noticeably change skin finish, realism, mood, and medium response, so expect a different combination rather than an identical result. Feel free to experiment. :3