<h1 id="flux.2-klein-9b-reference-based-face-swap-workflow">FLUX.2 Klein 9B - Reference-Based Face Swap Workflow</h1>⚡ Production-tested workflow.This workflow originated from an internal automation pipeline capable of processing hundreds of images automatically through the ComfyUI API on a single NVIDIA T4 GPU.A highly consistent face swap workflow for ComfyUI built around FLUX.2 Klein 9B, latent references, and a dedicated face swap LoRA.Unlike traditional face swap methods, this workflow uses both a Target Reference and a Source Face Reference to preserve composition while transferring identity.<hr /><h1 id="requirements">Requirements</h1>Before loading the workflow, make sure you have:<ul><li>FLUX.2 Klein 9B</li><li>FLUX2 VAE</li><li>Qwen 3 8B FP8 Mixed Text Encoder</li><li>Face Swap LoRA</li><li>LanPaint Custom Nodes</li></ul><hr /><h1 id="required-models">Required Models</h1><h2 id="flux.2-klein-9b">FLUX.2 Klein 9B</h2>Place inside:<pre><code>ComfyUI/models/unet/
</code></pre>Model:<pre><code>flux-2-klein-9b.safetensors
</code></pre><hr /><h2 id="flux2-vae">FLUX2 VAE</h2>Place inside:<pre><code>ComfyUI/models/vae/
</code></pre>Model:<pre><code>flux2-vae.safetensors
</code></pre><hr /><h2 id="qwen-3-8b-fp8-mixed">Qwen 3 8B FP8 Mixed</h2>Place inside:<pre><code>ComfyUI/models/text_encoders/
</code></pre>Model:<pre><code>qwen_3_8b_fp8mixed.safetensors
</code></pre><hr /><h2 id="face-swap-lora">Face Swap LoRA</h2>Place inside:<pre><code>ComfyUI/models/loras/
</code></pre>Model:<pre><code>bfs_head_v1_flux-klein_9b_step3500_rank128.safetensors
</code></pre><hr /><h1 id="required-custom-nodes">Required Custom Nodes</h1><h2 id="lanpaint">LanPaint</h2>This workflow uses the LanPaint sampler.Install it through:ComfyUI Manager → Install Missing Custom Nodes<hr /><h1 id="workflow-overview">Workflow Overview</h1>This workflow uses two different references:<h3 id="target-reference">Target Reference</h3>Preserves:<ul><li>Composition</li><li>Pose</li><li>Framing</li><li>Perspective</li><li>Scene structure</li><li>Lighting context</li></ul><h3 id="source-face-reference">Source Face Reference</h3>Transfers:<ul><li>Face identity</li><li>Facial proportions</li><li>Hair characteristics</li><li>Eyes</li><li>Skin details</li></ul>By combining both references, the workflow achieves significantly better consistency than traditional face swap approaches.<hr /><h1 id="how-to-use">How To Use</h1><h2 id="step-1-load-the-source-face">Step 1 - Load The Source Face</h2>Node:Source Face ImageThis image provides the identity that will be transferred.Recommended:✅ Front-facing portraits✅ High-resolution images✅ Clear facial visibility✅ Good lightingAvoid:❌ Motion blur❌ Heavy occlusions❌ Sunglasses❌ Extreme angles<hr /><h2 id="step-2-load-the-target-image">Step 2 - Load The Target Image</h2>Node:Target ImageThis image provides:<ul><li>Background</li><li>Body</li><li>Clothing</li><li>Pose</li><li>Composition</li><li>Framing</li><li>Perspective</li></ul>The workflow attempts to preserve these elements while replacing the identity.Recommended:✅ Portrait photos✅ Medium shots✅ Visible face✅ Clear lightingAvoid:❌ Tiny faces❌ Hidden faces❌ Extreme side profiles<hr /><h2 id="step-3-queue-prompt">Step 3 - Queue Prompt</h2>Press:Queue PromptThe workflow automatically:<ol><li>Reads target dimensions</li><li>Creates target latent references</li><li>Creates source face latent references</li><li>Applies dual-reference conditioning</li><li>Generates the face-swapped result</li><li>Preserves original scene composition</li></ol>No additional setup is required.<hr /><h1 id="what-this-workflow-preserves">What This Workflow Preserves</h1><ul><li>Original background</li><li>Original composition</li><li>Original lighting</li><li>Original perspective</li><li>Original pose</li><li>Original framing</li><li>Original camera distance</li><li>Original scene layout</li></ul><hr /><h1 id="ideal-use-cases">Ideal Use Cases</h1><ul><li>UGC Creator Replacement</li><li>Influencer Face Swaps</li><li>Character Consistency</li><li>Marketing Creatives</li><li>AI Avatars</li><li>Personal Branding</li><li>Dataset Generation</li><li>Batch Face Swap Pipelines</li></ul><hr /><h1 id="performance">Performance</h1>Tested on:<ul><li>NVIDIA T4</li><li>RTX 3060</li><li>RTX 4070</li><li>RTX 4090</li></ul>VRAM requirements will vary depending on image resolution.<hr /><h1 id="batch-automation">Batch Automation</h1>This public release contains the single-image workflow.Internally, I use a custom Python automation system connected directly to the ComfyUI API that can process large batches automatically.Example setup:<ul><li>1 source face</li><li>400+ target images</li><li>Automatic queue management</li><li>Automatic retries</li><li>Organized output folders</li><li>GPU-friendly processing</li><li>Designed for T4 cloud instances</li></ul>This automation system is not included in the public release.<hr /><h1 id="commercial-applications">Commercial Applications</h1>This workflow can be used for:<ul><li>UGC generation</li><li>Marketing campaigns</li><li>Creator replacement</li><li>Character consistency projects</li><li>AI influencer pipelines</li><li>Content automation</li></ul><hr /><h1 id="need-batch-processing">Need Batch Processing?</h1>I also develop custom automation systems built around ComfyUI.Examples:<ul><li>Face swap automation</li><li>Folder-based processing</li><li>ComfyUI API integrations</li><li>Large-scale image generation pipelines</li><li>Dataset creation workflows</li><li>Custom workflow development</li></ul>If you're interested in automation or custom solutions, feel free to send me a direct message on Civitai.<hr /><h1 id="credits">Credits</h1>Built and tested using FLUX.2 Klein 9B, latent reference conditioning, and a dedicated face swap LoRA for high-consistency identity transfer.If you find this workflow useful, consider leaving a rating and sharing your results.

FLUX.2 Klein 9B - Reference-Based Face Swap Workflow

Initial release.Reference-based face swap workflow for FLUX.2 Klein 9B using dual latent conditioning.Features:- Source face reference- Target composition reference- Face swap LoRA integration- LanPaint sampling- Automatic target size detection- High composition preservationDesigned for single-image face swaps and compatible with large-scale automation pipelines through the ComfyUI API.

Files with SHA256: d41a242aa61616bec1e1c66b02713487d67bc7f6e9c0d378cab468e86faea65c