Flux Identity Adjustor Node balances reference and prompt for consistency

The Flux Identity Adjustor Node is a new tool for the Flux.2 klein 9B model that focuses on maintaining identity consistency while generating images from prompts. The developer, a Reddit user with limited Python skills, built the node through ‘vibe coding’ – using three different AIs over 1.5 weeks to combine multiple techniques. The node essentially acts as a dynamic balancer between a reference image (e.g., a face) and the text prompt, adjusting the influence of each to produce outputs that preserve the subject’s identity without sacrificing creative variation.

Tested exclusively on the Flux.2 klein 9B FP8 distilled version using a standard ksampler on an RTX 2060, the developer notes that default identity blocks 8-19 are best for artistic themes. For skin texture issues, users can switch to blocks 6-15 or 8-16 as Flux processes texture in blocks 17-23. The node is open-source on GitHub and builds upon earlier work by another Reddit user. Early examples show Jason Statham in various scenes, indicating strong real-world potential for consistent character generation.