Custom DeepFaceLive RTM Model
Publicada el 2026-07-21
Descripción de la oferta
I need a single-identity DeepFaceLive model (.dfm) trained with the “Ready-to-Merge” workflow so it can be dropped onto any target video without retraining. The final file must cover the whole face at 224 px (320 px if it noticeably improves quality) and retain good likeness even on difficult profiles and dynamic, mixed-lighting shots—those angles are already well represented in my dataset of roughly 200-400 curated photos plus 30–50 short clips. Please start from a solid DeepFaceLab pre-train, then continue until the source loss plateaus and broad-angle fidelity is clear. When the model looks ready, run a quick swap on a test clip I will supply; that demo will serve as the milestone for approval and final release of funds. On delivery I expect: • The finished .dfm file, clearly named. • The short demo swap (H.264 or ProRes is fine). • Written confirmation that all of my source media has been permanently deleted. Let me know your typical turnaround time, how many tweak rounds you include, and attach or link to a few previous DeepFaceLive / face-swap samples so I can judge style and quality. The end use is entertainment and media production, so real-time performance and a clean, artifact-free result are priorities. Looking forward to your proposal—feel free to ask if any details are missing.
Skills
Fuente original: freelancer