FisherKing-Flux2.Klein-9B-[CharacterBlender]-ReferenceWorkflow-v1.0 [Low VRAM]

Design Philosophy

This workflow is intended as a reference implementation. It prioritizes clarity, modularity and low VRAM compatibility over automation, making it easy to understand, modify and extend.

Purpose

• Blend the identity of a Character Reference into an Environment Reference.

• Optimized for Flux2Klein GGUF.

• Low VRAM friendly.

Workflow Steps

1. Load Character Reference

2. Load Environment Reference

3. Adjust Positive Prompt

4. Adjust Character Mask Settings (Enable / Disable relevant switches)

5. Adjust Environment Mask settings (only if needed)

6. Queue Prompt

Notes

This workflow blends two reference images using Flux2Klein's reference conditioning. In general, the first reference has a stronger influence on identity while the second reference contributes composition and scene context, although the final blend depends on the compatibility of both references and the model's learned priors.

• Character identity comes primarily from Character Reference.

• Scene composition comes primarily from Environment Reference.

• Prompt refines the final result.

• Use high-quality references for best output.

Reference Selection

• This workflow is designed for character replacement by blending two reference images.

• Character Reference: The identity, facial features and appearance you want to preserve.

• Environment Reference: The composition, pose, camera angle, lighting and scene you want to inherit.

Best Results

• Use references with similar framing and aspect ratios.

• Similar poses and camera angles generally produce more natural replacements.

• Large differences in framing (Landscape vs Portrait), pose or composition may produce unexpected results.

Reference Images

• Reference images are automatically normalized to a maximum dimension of 1280 px.

• This reduces encoding time and VRAM usage while preserving visual fidelity.

• Increase the maximum resolution for higher reference fidelity if additional GPU memory is available.

Mask Configuration

• The default masking pipeline is optimized for realistic and near-realistic images.

• Stylized artwork (Anime, Manga, Cartoons, etc.) may require an alternative segmentation model or manual mask refinement.

• Depending on the reference image, you may need to adjust the face, hair or confidence settings for improved segmentation.

• Images with complex hairstyles, accessories or occlusions may require additional tuning.

• This workflow provides a validated baseline. Feel free to adapt the masking pipeline for your own datasets and use cases.