SilverScreen 110 applies the strongest push, past the reference model's own level (alpha = 1.10).

[Recipe] Add-difference merge on UNet only: PerfectDeliberate v10 + 1.10 x (Nova Reality XL IL v9.0 - Nova Anime XL v19). VAE and text encoders are inherited from PerfectDeliberate. Illustrious-family checkpoint, fp16 safetensors, no baked LoRA.

[Why add-difference] Plain weighted averaging only interpolates between two checkpoints, so on similar parents the four variants collapse into one look. The add-difference form extracts the realism direction (Nova Reality minus Nova Anime) and scales it, which keeps the four rungs visibly apart. Alpha above 1.0 extrapolates beyond the reference and is intentional here.

[Series] SilverScreen IL: 25 / 50 / 80 / 110 on the same base, one direction only - realism.

[Recommended settings] Euler a, 24-28 steps, CFG 5.0-5.5, 832x1216. Hires fix above 1.25x helps skin detail.

[Known limitations] Two-person prompts often collapse to a single subject. At alpha 1.10 contrast and micro-detail are pushed hard, so lower the CFG or step back to 80 if results look over-sharpened. East Asian facial priors dominate unless ethnicity is stated. Illustrious-family only.

[Parents and license] PerfectDeliberate v10: https://civarchive.com/models/24350 ; Nova Reality XL: https://civarchive.com/models/453428 ; Nova Anime XL (reference only): https://civarchive.com/models/376130 . All three permit derivatives and generation services (Rent). Selling this checkpoint or its merges is not granted by the parents. Parent credit retained.