Testing Minimax H3: Multi-Reference Prompts Ignored, 5-Hour Generation on RTX 4090
RikkTheGaijin77 · reddit · 2026-08-05
A user testing the multi-reference model of Minimax H3 via ComfyUI encountered several issues. Firstly, the model completely ignored the provided face close-up reference, retaining the face from the full-body image instead. Secondly, it disregarded the fixed-camera setup of the reference dance video and autonomously added zooms and close-up shots.
Furthermore, the generation speed raised questions. Generating a 15-second, 1MP video took 5 hours and 43 minutes on an RTX 4090 with 128GB of RAM, heavily contrasting with community claims of generating similar videos in minutes on an RTX 4060.
Related event: MiniMax H3 Local Tests: Fast but Flawed in Quality and Consistency(19 posts)→
More from Multimodal
- Kling 4.0 fully available, shows markedly better motion coherence in fast-action scenes — alifcoder · 2026-10-16
- 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