IP-Adapter at 0.6 overrides prompts; lower it and 4-view character consistency breaks

God_Speedmyboy · reddit · 2026-09-04

An indie dev generates chibi character turnaround sheets (front/3-4/side/back) for a mobile idle game using SDXL: Blender base mesh → depth maps from 4 angles → ControlNet Depth for pose → IP-Adapter style mode at 0.6 weight.

The dilemma: poses are perfect, but at 0.6 the reference 'bullies' the prompt — khaki pants and a hair flaw get copied into every view. Lowering the weight lets prompts through, but the 4 views stop matching. They ask whether to use attention masks, a different IP-Adapter mode, or train a style LoRA first.

Related event: Indie Dev Dilemma: IP-Adapter Weight vs. Multi-View Consistency in SDXL Sprites(2 posts)→

Original post →

More from Multimodal

Multimodal channel →