Feature Request: Per-Element ControlNet Strength Control in AI Studio (3D to Video)
I’ve been using AI Studio’s 3D to Video workflow to bring a fully built 3D scene (Victorian mansion, grounds, atmosphere) to life, driven by a first-frame style reference and a 3D viewport video capture, running through Seedance 2.0.
The environment results have been genuinely excellent — the architecture, lighting, and atmospheric effects (rain, mist, reflections) hold up beautifully and stay faithful to the reference frame.
The problem is the character. My scene includes a human actor (one of the premium AI Actors) walking through the environment. No matter how I adjust the prompt, the character consistently renders , artificial, and game character-like — a world away from the photorealistic human results I get when I prompt a person directly in freeform text-to-video with no 3D/ControlNet input driving it. The difference in quality between “AI generating a person freely” and “AI generating a person constrained by ControlNet motion/depth data” is enormous, and right now AI Studio gives no way to close that gap.
I understand AI Render (the ComfyUI plugin) already exposes exactly the control I need — per-input ControlNet strength and effect range sliders. That tells me the capability clearly exists in the underlying pipeline. What’s missing is any version of that control inside AI Studio itself, which is built as the simplified, non-technical front door for exactly the users (like me) who don’t want to build ComfyUI node graphs.
The request: even a single additional slider in AI Studio’s 3D to Video panel — something like “Character Fidelity” or “Actor ControlNet Strength” — that lets us loosen the model’s obligation to strictly follow a character’s rigged/driven motion, independent of the strength applied to the environment/architecture. Right now there’s one global process with no way to tell the system “hold the house exactly, but give the actor more room to move naturally.”
As more people push AI Studio toward narrative, character-driven short films rather than product demos or single-subject showcases, I suspect this gap is going to come up a lot. Right now, for any shot where a believable human performance matters, AI Studio’s simplicity is actually the thing standing between us and a usable result — and the fix (based on what AI Render already proves is possible) seems like it would be a UI addition more than a new engineering problem.
Happy to share test renders/examples if that’s useful for the team to see the issue directly.