Single-prompt music video: A Minimax H3 workflow breakdown
xI_AM_AFRICAx · reddit · 2026-08-30
A detailed breakdown of generating a music video using Minimax H3's Ref2VA workflow on a 3090 (64GB).
Workflow:
- Generated in 15-second increments with a 1-second overlap (e.g., second segment starts at 14s) for smooth stitching.
- Post-processed with Topaz Video for noise reduction and upscaling.
- Inputs: Two portrait images and one audio track from Suno 3.5.
Prompt Strategy:
- Used a single, highly detailed prompt for all generations, defining the subject's appearance (mullet, mustache, headset), audio syncing requirements, and studio lighting.
- Only modified the audio start/end duration parameters during generation.
Lessons Learned:
- Changing reference images during re-generation caused character inconsistency; consistency in references is key.
- H3 showed impressive lip-sync accuracy and even interpreted strumming motions to match the audio.
Related event: Suno Music Meets Minimax H3 for Lip-Sync Music Videos(4 posts)→
More from Multimodal
- FLUX 3 Image Klein lands on fal: box-drawn layouts, multi-reference placement, style transfer — bfl_ai · 2026-10-16
- SGLang ships Day-0 support for FLUX 3 Image Klein: 10 reference images, BF16/FP8/NVFP4 variants — ying11231 · 2026-10-16
- Krea 3 Scores 4.81 on CPI-Bench, Second Only to GPT Image 2.5 — testingcatalog · 2026-10-16
- Higgsfield teases Kling 4.0 rollout with a monkey-driven demo video — azed_ai · 2026-10-16
- Halcyon early-fusion image models match late-fusion with 275x less pretraining compute, open-sourced — davidwromero · 2026-10-16
- Running Wan2.1 14B FP16 on a 12GB laptop GPU: VRAM swap stress test finishes in 4m11s — RYuu-1120 · 2026-10-16