Vision Vanguard 25 [ERNIE-Image]
Film-grade documentary realism: material textures, natural light and cinematic shadows.
Base model: ERNIE-Image
Trigger words: none
Helpful style keywords: documentary photograph, cinematic realism, high-contrast lighting, shallow depth of field, textured concrete, clinical atmosphere
Strength: 0.8–1.0 (1.0 for the full style, 0.8 to blend more easily with your subject). The version specs list 0.85–1.0.
Prompting: write natural-language sentences, not keyword lists. A good order is subject → action/state → environment and materials → lighting → camera (for example "eye-level medium shot" or "macro close-up").
Tips
Works best with muted palettes plus one or two color accents.
Naming materials ("raw concrete", "weathered metal", "coarse fur") brings out texture detail.
Describe a scene, not just a subject.
Try harsh backlighting or chiaroscuro.
Description
Retrained with a lower learning rate, so it produces fewer artifacts and hallucinations than V2026.1, which was overfitted.
Base model: ERNIE-Image
Strength: 0.85–1.0
Resolution: trained at 512, generate at 1024
CFG: 4.6 (Base) / 1.0 (Turbo)
Steps: 35–50 (Base) / 8–9 (Turbo)