fix(sampler): the value function path steered without saying so

Only the fixed-direction fallback announced its lift, so a working value
function was indistinguishable from nothing happening.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

samplers.py

@@ -4894,6 +4894,14 @@ class FunPackLTXAVSceneChainSampler:
                 grad = value_fn.gradient(source)
                 unit = torch.nn.functional.normalize(grad.float(), dim=-1)
                 vf_state["d"] = self._direction_in_cond_space(model, raw_cond, unit, cond)
+                if vf_state["d"] is not None:
+                    # Say so out loud: the fixed-direction fallback announces itself when it
+                    # lifts, so silence here reads as "nothing happened" rather than success.
+                    lifted = int(vf_state["d"].shape[-1]) != int(unit.shape[-1])
+                    print("[FunPackSceneChain] embed_guidance: steering on the value function "
+                          "gradient" + (f", lifted {int(unit.shape[-1])} -> "
+                                        f"{int(vf_state['d'].shape[-1])} through the model's own "
+                                        "text preprocessor" if lifted else ""))
                 if vf_state["d"] is None:
                     print("[FunPackSceneChain] embed_guidance: the value function gradient "
                           "cannot be mapped onto this model's conditioning, using fixed direction")

tests/test_minimax_h3_sampler.py

@@ -884,3 +884,11 @@ def test_an_unusable_value_function_falls_back_to_the_fixed_direction(capsys):
     out = capsys.readouterr().out
     assert out.count("value function gradient failed") == 1   # once per run, not per step
     assert not torch.equal(seen["cond"], cond)                # fixed direction still steers
+
+
+def test_a_working_value_function_says_so_instead_of_steering_silently(capsys):
+    _run_embed_wrapper_with_vf(_h3ish_model(), torch.randn(1, 12, TEXT_DIM),
+                               torch.randn(1, 12, HIDDEN), _RawSpaceValueFunction(), steps=4)
+    out = capsys.readouterr().out
+    assert out.count("steering on the value function gradient") == 1
+    assert "5120 -> 5376" in out