Kling 1.6 vs Runway Gen-3 for consistent character identity

I've been spending the last week trying to get a consistent protagonist across a 15-second narrative sequence. My workflow involves a specific character sheet (front, profile, and 3/4 views) that I feed into the image-to-video models. I am currently torn between Kling 1.6 and Runway Gen-3 Alpha. Runway gives me better lighting control and more cinematic camera movements, but the character's facial features drift significantly after the second prompt iteration, especially around the eyes and jawline. Kling holds the face much tighter for the first few seconds, but it struggles with complex hand movements and often breaks down when the character interacts with props. I tried using the 'ControlNet' style approach in ComfyUI to stabilize the identity before feeding it to the video model, but that feels like a band-aid. Is there a specific parameter setting in either platform that actually locks the identity without freezing the motion? I've read that some people are using Lora adapters for this, but I haven't found a reliable generator that supports custom Loras for these newer video models. What is your current go-to if you need to maintain strict character consistency over a multi-shot sequence? Any advice on whether I should just stick to single-shot generations and edit them together in DaVinci would be appreciated. I don't want to spend another month testing random seeds.