Switch to the recommended ckpt850 EMA LoRA; default steps back to 4

ckpt850 is the final checkpoint of the round and sharp at 4 steps per
the updated card, so the UI/API default drops 8 -> 4. Also adds
H3_LORA_STRENGTH (default 1.0), the card's sharpness/artifact dial,
applied at fold time and in the per-request toggle.

README.md

@@ -47,12 +47,12 @@ callers get the same treatment the old upload event gave.
 ## 4-step Turbo LoRA
 
 The transformer runs with [`larryvrh/MiniMax-H3-Turbo-Lora`](https://huggingface.co/larryvrh/MiniMax-H3-Turbo-Lora)
-folded into its bf16 weights at startup (`h3_lora.py`), so the default is **8 sampling steps** instead of 28 (the
+folded into its bf16 weights at startup (`h3_lora.py`), so the default is **4 sampling steps** instead of 28 
 card's comfort zone at the current checkpoint; 4 is the design point but softer). The fold mirrors the diffusers key
 conversion exactly (fused-QKV thirds, the `SwiGLU` gate/value swap, the shared AdaLN row layout) and happens before
 the AoTI package is patched in, so compiled blocks carry the update too. The low-rank factors stay resident, so the
 **Turbo LoRA checkbox** (or the `use_lora` API parameter) folds/unfolds `lora_B @ lora_A` in place per request —
-off gives the base model at whatever step count you pick. `H3_LORA` selects the file (`off` disables entirely).
+off gives the base model at whatever step count you pick. `H3_LORA` selects the file (`off` disables entirely), `H3_LORA_STRENGTH` scales the update (the card's sharpness/artifact dial: >1 against smear, <1 against over-sharp grain).
 
 ## AoTI-compiled blocks
 
@@ -176,8 +176,9 @@ one-time `PIPE.to("cuda")` is inside the first row's 339 s and does not reappear
 | `H3_ATTENTION` | `_native_cudnn` | cuDNN's fused kernel, 10–20% faster than the SDPA default and needs nothing installed. flash-attention 3 is sm90-only and this pool is sm120. |
 | `H3_GPU_DURATION` | `900` | Seconds per request; the pool applies a 1.5 duration factor. |
 | `H3_GPU_SIZE` | `xlarge` | ZeroGPU allocation size. `large` does not fit. |
-| `H3_LORA` | `minimax_h3_turbo_4step_ckpt500.safetensors` | Turbo LoRA file folded into the transformer at startup. `off` disables. |
+| `H3_LORA` | `minimax_h3_turbo_4step_ema_ckpt850.safetensors` | Turbo LoRA file folded into the transformer at startup. `off` disables. |
 | `H3_LORA_REPO` | `larryvrh/MiniMax-H3-Turbo-Lora` | Hub repo the LoRA is fetched from. |
+| `H3_LORA_STRENGTH` | `1.0` | Scales the folded LoRA delta (sharpness/artifact trade-off). |
 
 ## Whose GPU quota pays
 

app.py

@@ -302,7 +302,7 @@ def _fit_keyframe(image_path, current_canvas):
     return image_path, label
 
 
-def generate(prompt, image_path=None, last_image_path=None, canvas=DEFAULT_CANVAS, duration=5, steps=8, seed=42, upsample=False, use_lora=True, ip_token=None):
+def generate(prompt, image_path=None, last_image_path=None, canvas=DEFAULT_CANVAS, duration=5, steps=4, seed=42, upsample=False, use_lora=True, ip_token=None):
     """One request. `upsample` is last and defaults off, so a positional API client that predates it is unaffected."""
     if LOAD_ERROR:
         raise Exception(LOAD_ERROR)
@@ -379,7 +379,7 @@ app = Server(title="MiniMax-H3 Studio")
 
 @app.api(name="generate")
 def _generate_api(prompt: str, image_path: FileData | None = None, last_image_path: FileData | None = None,
-                  canvas: str = DEFAULT_CANVAS, duration: float = 5, steps: int = 8, seed: float = 42,
+                  canvas: str = DEFAULT_CANVAS, duration: float = 5, steps: int = 4, seed: float = 42,
                   upsample: bool = False, use_lora: bool = True, request: Request = None) -> tuple[FileData, str, str]:
     """Generate a video with a synchronized soundtrack. Returns (video, report, refined prompt)."""
     # `request` is injected by the event system, not an API input; its x-ip-token bills the conditioner to the caller.

h3_lora.py

@@ -27,8 +27,11 @@ import os
 import torch
 
 LORA_REPO = os.environ.get("H3_LORA_REPO", "larryvrh/MiniMax-H3-Turbo-Lora")
-# The recommended default per the model card: ckpt500, non-EMA (newest, sharpest). `off` disables.
-LORA_FILE = os.environ.get("H3_LORA", "minimax_h3_turbo_4step_ckpt500.safetensors")
+# The recommended default per the model card: ckpt850 EMA (final checkpoint of the round, sharp at 4 steps).
+# `off` disables.
+LORA_FILE = os.environ.get("H3_LORA", "minimax_h3_turbo_4step_ema_ckpt850.safetensors")
+# The card's sharpness/artifact dial: >1 against blurry ghosting/smear, <1 against over-sharp grain.
+LORA_STRENGTH = float(os.environ.get("H3_LORA_STRENGTH", "1.0"))
 
 
 def _delta_targets(name: str, delta: torch.Tensor, inner_dim: int) -> list[tuple[str, torch.Tensor]]:
@@ -79,7 +82,7 @@ def apply_lora(transformer) -> str | None:
     for name in bases:
         a = lora[f"{name}.lora_A.weight"].float()
         b = lora[f"{name}.lora_B.weight"].float()
-        delta = b @ a  # alpha == rank, so the scale is 1
+        delta = (b @ a) * LORA_STRENGTH  # alpha == rank, so the base scale is 1
         targets = _delta_targets(name, delta, inner_dim)
         for key, converted in targets:
             param = params.get(key)
@@ -94,12 +97,14 @@ def apply_lora(transformer) -> str | None:
     transformer._lora_state = {
         "enabled": True,
         "inner_dim": inner_dim,
+        "strength": LORA_STRENGTH,
         "entries": [
             (name, lora[f"{name}.lora_A.weight"], lora[f"{name}.lora_B.weight"])
             for name in entries
         ],
     }
-    return f"LoRA `{LORA_REPO}/{LORA_FILE}` folded into {folded} weights ({len(bases)} modules)"
+    strength = "" if LORA_STRENGTH == 1.0 else f" @ strength {LORA_STRENGTH}"
+    return f"LoRA `{LORA_REPO}/{LORA_FILE}`{strength} folded into {folded} weights ({len(bases)} modules)"
 
 
 def set_enabled(transformer, enabled: bool) -> bool:
@@ -109,7 +114,7 @@ def set_enabled(transformer, enabled: bool) -> bool:
         return state["enabled"] if state else False
     import torch
 
-    sign = 1.0 if enabled else -1.0
+    sign = (1.0 if enabled else -1.0) * state.get("strength", 1.0)
     params = dict(transformer.named_parameters())
     for name, a, b in state["entries"]:
         delta = b.to(torch.float32) @ a.to(torch.float32)

index.html

@@ -211,8 +211,8 @@
           <input type="range" id="duration" min="2" max="14" step="1" value="5">
         </div>
         <div>
-          <div class="slider-row"><span>Steps</span><output id="steps-out">8</output></div>
-          <input type="range" id="steps" min="2" max="40" step="1" value="8">
+          <div class="slider-row"><span>Steps</span><output id="steps-out">4</output></div>
+          <input type="range" id="steps" min="2" max="40" step="1" value="4">
         </div>
         <div>
           <div class="deck-label">Seed</div>