Скачать или смотреть Getting Better Control in AI Video Generation with ComfyUI & Rhino

In this video, I dive into a new AI workflow using Wan Vace (14B Q5) and local video processing through ComfyUI—pushing my previous Rhino geometry further into high-quality architectural image generation.

⚙️ Workflow Breakdown:
– Captured Rhino geometry via screen recording
– Used it as a reference video in ComfyUI
– Combined with a static reference image
– Tested ControlNet with Depth (meh) and Canny (much better!)

🔥 These results are from the first attempts—not cherry-picked. Each generation took around 10 minutes, and the output is surprisingly coherent, especially in how reflections and side details show up in motion.

🧠 What’s Working:
The output feels like you're navigating a stylized 3D model. The composition, lighting, and transitions provide powerful feedback for early-stage design.

⚠️ Still Tricky:
Control remains a challenge. I'm now experimenting with masks to better balance between video input and reference image.

👇 Got a better method for balancing reference video vs image prompts? Drop it in the comments!

📌 Stay tuned for a full walkthrough—coming soon.