MiniMax H3 - Really is the best open-source video model.
Polite and respectful feedback is fine but this video turned out exactly as I planned.
Proof1 community thread
Key points
- • The song, made with Suno, lyrics created by Chat GPT.
- Model settings: MiniMax H3 Singularity, 25 steps on pass 1, then the 8-step 768p turbo LoRa at 4 steps and 0.35 denoise on pass 2 with the latent upscale model.
- • Planned all the scenes and wrote every H3 prompt shot by shot, with per-shot lyric cues and emotion tags
- We are working on a new story mode which will allow you to create video's just like this one when it comes to the story flow - But no need for Claude!
Sources (1)
- [1]MiniMax H3 - Really is the best open-source video model.r/StableDiffusion (top, day) · Oct 20, 03:36 AM
Polite and respectful feedback is fine but this video turned out exactly as I planned.
• The song, made with Suno, lyrics created by Chat GPT.
Extractive summary: sentences quoted from the sources.
Hype check
Independently evaluated8 independent looks since
- Evaluation · Oct 20, 2026DuplexLedger: Benchmarking Transactional Memory in Full-Duplex Speech Agents
- Evaluation · Oct 20, 2026Auditing the Product, Not the Recursion: The Evaluation Evidence Behind Recursive Self-Improvement Claims
- Evaluation · Oct 20, 2026CUA-Recovery: Benchmarking Error Recovery in Long-Horizon Computer-Use Agents
- Evaluation · Oct 20, 2026Corpus Quality Improves, Retrieval Utility Does Not: An Ablation of Preprocessing and Refinement in a Speech-to-QA Pipeline
- Evaluation · Oct 20, 2026Reusing Multimodal Agents: Efficient Evaluation of Sound Designs in Games
- Evaluation · Oct 20, 2026Memory as Code: Declarative Retrieval over Codified Long-Term Memory
Before this
- Oct 19, 2026EGI-BENCH: An Enterprise General Intelligence Benchmark Suite
- Oct 19, 2026Always On - A collaboration between two LLMs
- Oct 19, 202611 all-in-one ComfyUI nodes I built for my own 16GB VRAM card (RTX 5060 Ti)
- Oct 18, 2026Comfy-Org/Fibo-2.0
- Oct 18, 2026Elevatormusic/DeepSeek-V4.1-Flash-1017-EXL3-TensorFold-2x-DGX-Spark: DeepSeek-V4.1-Flash-1017 EXL3 2.9bpw on 2x NVIDIA DGX Spark (GB10) with TensorFold: 2.1-2.3x the vLLM kit's decode speed.
- Oct 17, 2026Blown away by Minimax H3 sometimes. It nailed this one!!!!
