Treating Minimax H3 as a 'Dumb Cameraman' to Stitch a 2-Minute Dialogue Scene
niechta · reddit · 2026-08-09
A developer shared a complete workflow for creating a 2-minute dialogue scene using the Minimax H3 model. To avoid single-prompt failures, the author treated the AI like a "dumb camera operator," generating separate takes, coverage shots, and silent reaction handles, then manually editing them together.
Core Tech Stack & Parameters:
- Model & Plugins: Minimax H3 reference-driven path (Ref2VA), using ComfyUI 0.30 with ComfyUI-H3-Multishot and ComfyUI-GGUF nodes.
- Specs: 960×540 resolution, 24 fps; the model generates picture and 32 kHz stereo audio in a single pass.
- Inference: 20 steps, cfg 1.0, resmultistep / simple sampler, fixed seed.
- Hardware: 420 seconds per 362-frame generation on an RTX PRO 6000.
The author notes that the resolution limitation causes noticeable pixelation in cropped single shots, but post-processing successfully glued all chunks into a cohesive piece.
More from Multimodal
- MJev-Spatial: Single-Pass Classification Beats Autoregressive VLMs on Spatial Reasoning — mindlab-research · 2026-10-16
- H-DDCM: Hadamard Codebooks Make Diffusion-Based Image and Video Compression Scalable — Technion · 2026-10-16
- ComfyUI v0.41.0 ships with Flux3 image model and MiniMax-H3 LightVAE support — Fcking_Chuck · 2026-10-16
- AI-generated short film 'The Dragon Rider' glides cinematically above the clouds — jijo6686 · 2026-10-16
- Realistic candid Midjourney portrait prompt: natural skin, lived-in bedroom light — tisch_eins · 2026-10-16
- 1986 to 2026 in one session: outpaint, modernize and 4x upscale, session file included — juanpablogc · 2026-10-16