https://www.kelvindesigns.tips/visual-ai-quickstart

built this today workflow 1.created retro mascot moodboard 2. prompted until got suitable geometry and style 3. took favorite as omni-ref to refine style 4. placed favorite in Nano for accuracy edits (added a specific weighted vest, fixed American flag) 5.will put in Photoshop+ illustrator next to customize the message and get print production ready

https://youtu.be/AUT3r6inKAk?is=kiBkc4xdXdQAYqly

https://www.loom.com/share/e50775ea312b4b81af4f997deaaba32a Hi all. Im trying to recreate a battlefield cross (military memorial subject) with dramatic lighting and a specific angle (2 boots, dog tag necklace, M4 rifle, combat helmet). I want to lock in a solid 3D-looking base image and then apply different styles and moods to it without the composition falling apart. Here's my workflow so far: - Generated a 3D model of the subject using Rodin (text-to-3D), took a screenshot at the angle I wanted, and used that as my --oref input - Tried varying --ow weights extensively — didn't help much - Used --sref codes on their own and combined with seeds....composition still drifts wildly - combined an image reference + style reference of the same subject I got close, but the moment I run it through Style Creator, everything shifts - Took edits into NanoBanana for inpainting, and manually adjusted things in Photoshop, but that breaks the iterative workflow I need - Currently experimenting with Weavy (node-based workflow combining multiple models) to see if I can get a more stable base Has anyone cracked a reliable way to lock composition when it's a "custom" composition and not a single subject (I'm assuming MJ is seeing this as possibly 5 subjects: 2 boots, dog tag necklace, M4 rifle, combat helmet)?