zjourney - Fantasy Realism Refiner | Krea 2 Update
Focused on further grounding Fantasy outputs in photographic "realism". Trained on 44 film stills.
Usage:
Recommended strength: 0.8-1.4
Examples generated with filter bypass + zjourneykrea LoRA 
A lightweight LoRA trained on 44 curated images generated with the MidjourneyV1 Z Image Turbo LoRA, built to refine Ideogram 4 fantasy outputs toward a more realism aesthetic.
Why I made this: Ideogram 4 produces stunning fantasy art, but that's often the problem. No matter how I prompted it (photographic, photorealism, cinematic realism, etc.), fantasy subject matter almost always defaults to an illustrated or painterly art style. zjourney was built to push past that and ground those outputs in something that feels genuinely photographic.
What it does: Refines Ideogram 4 outputs with richer photorealism, improving surface texturing, cinematic lighting, environmental depth, and material detail across fantasy and action subjects. It also works well as a general realism refiner beyond pure fantasy content.
zjourney - Fantasy Realism Refiner | v2 Update
V2 expands the training dataset, focusing on further grounding outputs in photographic realism. Trained on 44 film stills alongside 6 MidjourneyV1 Z Image Turbo LoRA generations.
Usage:
Standalone recommended strength: 0.55 0.35-0.6
For best results, stack with v1:
v1 @ 0.35 + v2 @ 0.45
First 2 example images generated standalone with v2
Final 5 example images generated with v1 + v2 combined
Key improvements over base Ideogram 4:
More detailed and grounded environments (ruins, foliage, wet stone)
Richer skin, fur, and fabric textures
Stronger cinematic color grading and atmospheric depth
More convincing crowd and secondary figure detail
Known limitations:
Close-up shots with heavy focus on skin texture can see a slight degradation in quality, works best at mid-range and wider compositions
Usage:
Recommended strength: 0.4–0.7
Sample images generated at 0.5
Works best alongside detailed fantasy or action prompts; can also be applied to general realism scenes
Works with and without trigger word, the trigger word was trained in the aesthetics placement in the json prompt structure