Grey-box Blender previs plus MiniMax H3 locally renders a 95-second action scene
Longjumping-Raise900 · reddit · 2026-11-23
The author turned a 95-second car-park action scene into live-action footage on a local PC using a grey-box previs → video model workflow:
Workflow
- Built the scene in Blender (Eevee, 24fps) as grey boxes: one distinct color per person, red block for the car, with camera moves, cuts and timing locked in the previs.
- In ComfyUI (Easy-Media multitrack), the previs sits on the video track; H3 gets its slice as <Video 1> plus up to 9 reference images (character sheets, cars, one car-park plate).
- Split at shot cuts into 8 segments of 10-13s.
Setup & cost: RTX 4080 SUPER 16GB, 64GB RAM, 1.0 MP, 8 steps, single sampling mode, 20 min per segment, 3 hours for the film plus 3 retakes.
What worked / what didn't
- Worked: segment timing locked exactly to previs; wide shots closely match camera height and blocking; character sheets keep consistency across all 8 segments; box colors are an effective way to identify characters.
- Pitfalls: H3 treats previs as a guide, adding its own cuts and angles; one take added 20 extra bystanders — fixed by naming every person in the prompt and stating "the car park is empty, no crowd"; fast car shots are weakest with motion blur.
- The key prompt trick: declare that <Video 1> is the grey-box previs of this exact sequence — follow it shot-by-shot for camera, framing, every cut, timing and positions, while ignoring the boxes' colors/materials/lighting and rendering real people, a real car and a real car park.
More from Multimodal
- Dev builds interactive 3D human eye with Gemini Neon to teach how vision works — Scobleizer · 2026-11-23
- Princeton study: diffusion models barely extrapolate basic visual skills, ~50% on counts — NielsRogge · 2026-11-23
- AI drops Goku into real life with convincing physics and lighting, blurring what's real — heypearlai · 2026-11-23
- FaceRefine Plus v0.2.0 speeds up MiniMax H3 video face refining by 2.7x — AnniCodex · 2026-11-23
- Anima Artifacts demos a 'living magazine' you converse with instead of read — alexcovo_eth · 2026-11-23
- Turning real F1 telemetry into an AI video: 5 nights of failed experiments — BittiAI · 2026-11-23