MiniMax H3 day-one test finds better identity retention with max ref image size
jozbgm · reddit · 2026-08-04
After one day with MiniMax H3, the author says they’ve basically spent the entire day testing it and building from a real benchmark clip.
What they tested
- Recreated a short clip from an earlier Seedance video using H3.
- Kept the same character and beats, stitched from 6-second clips into a 42-second sequence.
- Used ComfyUI, an RTX 5090, 96GB RAM, and the minimaxh3ref2vaprunedint8convrot workflow.
What seems to help
- refimagesize: max works better than matching size when faces or textures must persist.
- resmultistep + beta scheduler seems better for reference-heavy prompts.
- Describe sound as physical events in time—impact, tear, breath—instead of mood words.
- Re-anchor the character explicitly in every clip rather than assuming continuity.
Open questions
- How much motion can fit in one clip.
- How well identity survives across long chains of generations.
- Whether some cuts should be generated in one shot instead of stitched later.
The author says they’re only at hour one of the optimization curve and invites others to share settings and prompt habits.
Related event: MiniMax H3 Local Tests: Benchmark Results and Comparisons Across GPUs(88 posts)→
More from Infra
- China's AI boom drives up to 660% profit surges for domestic chip firms; GPU maker MetaX turns profitable — pstAsiatech · 2026-10-16
- Qdrant tests Perplexity's visual PDF embeddings: 21x smaller vectors, 99.9% after rescoring — jrxna · 2026-10-16
- When Does an Agent Actually Need an API Gateway? Reddit Weighs Direct Calls vs. Unified Endpoints — TurbulentStiffness · 2026-10-16
- Taiwan to build $4B AI manufacturing complex in Texas to back Nvidia and TSMC — Polymarket · 2026-10-16
- When the KV Cache Is Slower Than the GPU: Remote KV Tiers Backfire in Production — MultiverseComputingCAI · 2026-10-16
- GitHub rewrote the 800k-line Copilot runtime in Rust using AI agents, cutting CPU and memory by over 90% — bibryam · 2026-10-16