AI Video Face Swap
Create consented deepfake-style video face swaps with your fal.ai key and provider billing. Klopaper keeps job metadata, not your media files.
Close, well-lit framing with the face visible through the shot.
Recommended for identity consistency and facial motion.
Generation controls
Upload a guide and reference face, describe the identity and action, then run.
How V3 works
Identity stays in view throughout diffusion
Klopaper builds the model’s green side-strip guide locally on the server, places your reference face in every frame, and sends that temporary composite to fal.ai. After generation it removes the strip and uploads the clean video back to your fal storage.
- Use a short, well-lit clip with one clearly visible face.
- Describe identity only in Face. Describe movement, clothing, and scene only in Action.
- Compare V3 Adaptive and Rank 64 when a shot drifts or deforms.
Known limits from the model author: hard cuts can reset identity, wide shots reduce fidelity, and preparation quality matters more than pushing LoRA strength.
Model and workflow by Alissonerdx. Inference uses fal.ai LTX 2.3 video-to-video with LoRA. Research, VFX, and consented identity work only.
Need a still image instead? Switch to the AI photo face swap tool for single images or batch combinations with Qwen Image Edit.