Add final performance benchmarks + hosted inference link

README.md

@@ -12,6 +12,9 @@ tags:
 
 # MiniMax H3 ref2va — NVFP4 (ComfyUI-native)
 
+> 🚀 **Use this model hosted (API + playground):**
+> [modelslab.com/models/minimax/minimax-hailuo03-reference-to-video](https://modelslab.com/models/minimax/minimax-hailuo03-reference-to-video)
+
 NVFP4 quantization of the [MiniMax H3](https://huggingface.co/MiniMaxAI/MiniMax-H3)
 **reference-to-video** DiT (`ref2va`), in ComfyUI's native quant layout — loads with
 the stock `UNETLoader` on any Blackwell GPU (sm_120: RTX 5090 / RTX PRO 6000, and
@@ -57,8 +60,24 @@ the FP4 path runs emulated and is ~2.3× slower).
 | **nvfp4_fpmm (this repo)** | 38.6 GB | — | 14.99 | ties bf16 on cu128, 42 % less VRAM |
 | pruned int8_convrot (official) | 21.0 GB | — | 22.7 | pruned arch, W8A16 dequant path |
 
-Stacking ComfyUI's `EasyCache` (threshold 0.1) on top cuts wall time a further
-~1.4× with deviation within the pipeline's run-to-run variance.
+Stacking ComfyUI's `EasyCache` (threshold 0.2, quality-gated) cuts wall time a
+further ~1.4×.
+
+## End-to-end speed (full optimized stack)
+
+600W RTX PRO 6000 Blackwell, `nvfp4` + cu130 + SageAttention +
+fp16_accumulation + fused adaln/gate kernels + EasyCache 0.2, 20 steps,
+1344×768 @ 24 fps with native stereo audio:
+
+| Video length | Wall time | Notes |
+|---|---|---|
+| **15 s** (362 frames — the model's one-shot ceiling) | **8 min 07 s** | vs ~25 min on the stock template path (~3×) |
+| **5 s** (124 frames) | **~2 min** (hot server) | measured through a production API |
+| 15 s draft (16 steps @ 960×544) | ~3.5–4 min | preview tier |
+
+Important: H3's trained range tops out at 362 frames (~15 s) — longer one-shot
+generations collapse regardless of settings (verified empirically at 736
+frames with multiple schedules/shifts).
 
 Same-seed visual quality of `nvfp4` vs bf16: no quantization artifacts observed
 (trajectory divergence only — diffusion is chaotic under any weight