diff --git a/modules/models/ltx/pipeline.py b/modules/models/ltx/pipeline.py
index 529a8ab5..81924131 100644
--- a/modules/models/ltx/pipeline.py
+++ b/modules/models/ltx/pipeline.py
@@ -16,7 +16,7 @@ the conditioning, the sound's length and the saved video agree on.
 DISTILLED_SIGMAS = "1.0, 0.99375, 0.9875, 0.98125, 0.975, 0.909375, 0.725, 0.421875, 0.0"
 
 
-def _base(image_to_video: bool):
+def _base(image_to_video: bool, as_guide: bool = False):
     video = {"at": "project.video", "input": "frame_rate", "label": "FPS", "drives": "fps"}
     latent_source = ["latent", 0]
     model_source, positive_source, negative_source = ["modifiers", 0], ["conditioning", 0], ["conditioning", 1]
@@ -51,13 +51,29 @@ def _base(image_to_video: bool):
              "inputs": {"media_id": ""}},
             {"id": "split", "group": "Preparation", "node": "LTXVSeparateAVLatent",
              "inputs": {"av_latent": ["latent", 0]}},
-            {"id": "anchor", "group": "Preparation", "node": "LTXVImgToVideoInplace",
-             "inputs": {"vae": ["model", 2], "image": ["source_image", 0], "latent": ["split", 0],
-                        "strength": 1.0, "bypass": False}},
-            {"id": "join", "group": "Preparation", "node": "LTXVConcatAVLatent",
-             "inputs": {"video_latent": ["anchor", 0], "audio_latent": ["split", 1]}},
         ]
+        if as_guide:
+            # The picture steers through guide attention at frame 0 and is NOT the clip's first frame:
+            # the latent stays empty, so the opening is free to move. Strength is the slot's own number.
+            slots += [
+                {"id": "anchor", "group": "Preparation", "node": "LTXVAddGuide",
+                 "inputs": {"positive": ["conditioning", 0], "negative": ["conditioning", 1], "vae": ["model", 2],
+                            "latent": ["split", 0], "image": ["source_image", 0], "frame_idx": 0, "strength": 0.35}},
+                {"id": "join", "group": "Preparation", "node": "LTXVConcatAVLatent",
+                 "inputs": {"video_latent": ["anchor", 2], "audio_latent": ["split", 1]}},
+            ]
+            guided = (["anchor", 0], ["anchor", 1])
+        else:
+            slots += [
+                {"id": "anchor", "group": "Preparation", "node": "LTXVImgToVideoInplace",
+                 "inputs": {"vae": ["model", 2], "image": ["source_image", 0], "latent": ["split", 0],
+                            "strength": 1.0, "bypass": False}},
+                {"id": "join", "group": "Preparation", "node": "LTXVConcatAVLatent",
+                 "inputs": {"video_latent": ["anchor", 0], "audio_latent": ["split", 1]}},
+            ]
+            guided = (["conditioning", 0], ["conditioning", 1])
         latent_source = ["join", 0]
+        positive_source, negative_source = guided
     slots += [
         {"id": "settings", "group": "Preparation", "node": "FunPackModifierSettings", "inputs": {"settings": "{}"}},
         {"id": "modifiers", "group": "Preparation", "node": "FunPackLoadModifiers",
@@ -68,8 +84,8 @@ def _base(image_to_video: bool):
             # A reference face carried into the clip. The picture is the scene's first reference
             # (assets.reference_1): with none picked this node passes everything through untouched.
             {"id": "identity", "group": "Preparation", "node": "FunPackIdentityTransfer",
-             "inputs": {"model": ["modifiers", 0], "vae": ["model", 2], "positive": ["conditioning", 0],
-                        "negative": ["conditioning", 1], "latent": latent_source, "source_id": 2.0,
+             "inputs": {"model": ["modifiers", 0], "vae": ["model", 2], "positive": positive_source,
+                        "negative": negative_source, "latent": latent_source, "source_id": 2.0,
                         "phase_scale": 1.0, "id_strength": 1.0, "arcface_mode": "auto_adjust"}},
             {"id": "identity_source", "group": "Reference media", "node": "FunPackLoadMedia",
              "roles": [{"at": "assets.reference_1", "input": "media_id",
@@ -87,8 +103,9 @@ def _base(image_to_video: bool):
                     "seed": 0, "steps": 8, "cfg": 1.0, "sampler_name": "euler", "scheduler": "normal",
                     "denoise": 1.0}},
 
+        *_uncrop(as_guide),
         {"id": "decode", "group": "Render", "node": "FunPackDecode", "inputs": {
-            "samples": ["sampler", 0], "vae": ["model", 2], "model": ["model", 0], "audio_vae": ["audio_vae", 0]}},
+            "samples": ["rejoin", 0] if as_guide else ["sampler", 0], "vae": ["model", 2], "model": ["model", 0], "audio_vae": ["audio_vae", 0]}},
         {"id": "video", "group": "Render", "node": "CreateVideo",
          "roles": [{"at": "project.video", "input": "fps", "label": "FPS", "drives": "fps"}],
          "inputs": {"images": ["decode", 0], "fps": 25.0, "audio": ["decode", 1]}},
@@ -98,9 +115,24 @@ def _base(image_to_video: bool):
     return slots
 
 
+def _uncrop(as_guide):
+    """A guide sits in the latent as extra frames: take them off again before the picture is decoded."""
+    if not as_guide:
+        return []
+    return [
+        {"id": "unjoin", "group": "Render", "node": "LTXVSeparateAVLatent", "inputs": {"av_latent": ["sampler", 0]}},
+        {"id": "crop", "group": "Render", "node": "LTXVCropGuides",
+         "inputs": {"positive": ["identity", 1], "negative": ["identity", 2], "latent": ["unjoin", 0]}},
+        {"id": "rejoin", "group": "Render", "node": "LTXVConcatAVLatent",
+         "inputs": {"video_latent": ["crop", 2], "audio_latent": ["unjoin", 1]}},
+    ]
+
+
 def presets():
     return [{"id": "ltx23_text_to_video", "title": "LTX 2.3 · Text to Video", "slots": _base(False)},
-            {"id": "ltx23_image_to_video", "title": "LTX 2.3 · Image to Video", "slots": _base(True)}]
+            {"id": "ltx23_image_to_video", "title": "LTX 2.3 · Image to Video", "slots": _base(True)},
+            {"id": "ltx23_anchor_guide", "title": "LTX 2.3 · Anchor as guide (image steers, opening stays free)",
+             "slots": _base(True, True)}]
 
 
 PROVIDES = {"pipeline_presets": presets}
diff --git a/modules/models/ltx/tests/test_ltx.py b/modules/models/ltx/tests/test_ltx.py
index eca43e35..54894932 100644
--- a/modules/models/ltx/tests/test_ltx.py
+++ b/modules/models/ltx/tests/test_ltx.py
@@ -122,7 +122,7 @@ def test_the_empty_latent_node_passes_the_audio_vae_to_a_provider_that_asks_and_
     assert seen == {"plain": True, "wants": ("A", 30.0)}
 
 
-@pytest.mark.parametrize("pid", ["ltx23_text_to_video", "ltx23_image_to_video"])
+@pytest.mark.parametrize("pid", ["ltx23_text_to_video", "ltx23_image_to_video", "ltx23_anchor_guide"])
 def test_the_pipelines_build_a_real_graph(registered, pid):
     from core import graph
     from modules.models.ltx.pipeline import presets
@@ -143,7 +143,7 @@ def test_the_pipelines_build_a_real_graph(registered, pid):
 def test_the_presets_are_offered_through_the_registry(registered):
     from core import routes
     found = {preset["id"] for _spec, make in routes.modules().providers("pipeline_presets") for preset in make()}
-    assert {"ltx23_text_to_video", "ltx23_image_to_video"} <= found
+    assert {"ltx23_text_to_video", "ltx23_image_to_video", "ltx23_anchor_guide"} <= found
 
 
 def test_the_picture_is_found_in_both_shapes_a_sampling_latent_takes(comfyui):