Face swapping workflows increasingly demand not just a pasted face but consistent skin tone, lighting, pose, and hair/body blending. BFS packages a practical set of LoRA weights and curated ComfyUI/ workflow examples so creators and experimenters can run face/head/body replacements across several modern image diffusion families without rebuilding bespoke pipelines.
Key Capabilities
- Multi-family LoRAs: released weights target Qwen Image 2.1, Qwen Image Edit (2509/2511), Flux 2 Klein (4b/9b) and Krea 2, plus experimental LTX‑2 variants, letting you pick the weight that matches your base model.
- Granular swap levels: separate files for face-only, full head, head-with-hair blending and experimental full-body replacement, so you can choose how much of the reference to transfer.
- Input-order convention: several head releases intentionally invert image order (body first, face second); following the repository’s input-order rules materially improves alignment.
- Practical defaults and workflows: recommended starting LoRA strength is 1.0; the repo includes ComfyUI workflows, example images, and guidance on resolution/scale to reproduce results consistently.
- Safety guidance: the maintainer explicitly advises against using or sharing results of public figures or without consent.
Who It's For and Trade-offs
Great fit if you are an image-editing practitioner or hobbyist who: wants quick integration into existing ComfyUI/diffusers pipelines; needs different granularities of replacement (face/head/body); and can match a LoRA to its base model. Expect the best results when the LoRA and base model families align and when you follow the repo’s input-order and scaling recommendations. Look elsewhere if you need guaranteed pose-preserving full-body transfers in all cases (full-body is experimental and can mismatch pose/clothing), or if you lack the compute/setup to run the recommended base models and ComfyUI workflows. Also weigh legal and ethical constraints—these weights enable high-fidelity swaps that can produce non-consensual or deceptive content if misused.
Where It Fits
BFS is a practical, workflow-oriented collection of LoRAs for face/head/body swaps inside modern image-diffusion editing stacks. Compared with end-to-end specialized deepfake pipelines, BFS emphasizes modularity (LoRA + matching base model + workflow) and flexibility across multiple diffusion families rather than a single-purpose closed tool.
