Workflow Sharing: Building a Consistent AI Character with Multimodal Tools
Gocciole · reddit · 2026-08-09
The author shares a workflow for maintaining the visual consistency of an AI character across multiple images and videos, successfully applied to a virtual cosplay series.
Core Toolchain:
- Google Flow: Used for complex costume and attire transfer.
- Krea 2 Turbo: Generates remaining stills using an identity LoRA and a locked reference image.
- MiniMax H3 Ref2VA: Used for video generation with a two-reference structure (one image controls scene and lighting, the other locks facial identity), preventing face drifting during scene changes.
Prompting Technique: Employs a "KEEP-first" prompt strategy. Instead of writing fresh descriptions per shot, an anchor image is generated first. Subsequent prompts state what stays identical (face, makeup, outfit, room) before describing what changes.
More from Multimodal
- Generating a 90s Sci-Fi Silent Film with Grok Imagine: A Century-Spanning Epic — tetsuoai · 2026-08-09
- Tencent Hunyuan Announces WorldClaw 3D Generation Project — Uncle___Marty · 2026-08-09
- Seedance 2.5 Tested: Features Local Editing and Up to 50 Reference Assets — alifcoder · 2026-08-09
- ByteDance's Seedance 2.5 Hits Lumina Platform, Exceeds Expectations in Tests — alifcoder · 2026-08-09
- Mixing Midjourney and ChatGPT to Create Cinematic Miniature Horror Scenes — umesh_ai · 2026-08-09
- AI Filmmaking Isn't Just a Prompt: Creator Shows Complex Manual Editing Timeline — gen_ericai · 2026-08-09