Add LTX audio workflow + Text (Prompt) node

New workflow/workflow_ltx_audio_api.json: Audio Wave + Segments + LoadImage (subject)
+ Text (Prompt) holding the LTX audio system prompt -> Judge in chat mode
(json_output, model 3.5-9b) -> LTX beats JSON on the analysis output. Added a tiny
CalibratorText node ('Text (Prompt)', STRING source) needed to feed the judge's
forceInput system_prompt/user_prompt/axes sockets. README points to the workflow.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

README.md

@@ -101,6 +101,13 @@ seg1: slow dreamy intro          # by segment number (also "1:" or "S1:")
 cinematic, moody grade           # no prefix = global (applies throughout)
 ```
 
+**Ready-made workflow:** `workflow/workflow_ltx_audio_api.json` wires it all up —
+`Audio Wave + Segments` + `LoadImage` (subject) + `Text (Prompt)` (the LTX system prompt
+from `prompts/ltx_audio_relay_system_prompt.txt`) → the judge in chat mode
+(`json_output=true`, model `3.5-9b`). Load it, upload your audio + set your subject image,
+and the judge's `analysis` output is the LTX beats JSON. (`Text (Prompt)` is a small STRING
+node this pack adds to feed the judge's `system_prompt`/`user_prompt`/`axes` sockets.)
+
 ### Interactive: `Audio Wave + Segments`
 
 Same outputs, but with an in-node waveform widget: **upload** an audio clip, **play** it,

__init__.py

@@ -16,9 +16,15 @@ from .nodes.audio_wave_segments import (
     NODE_CLASS_MAPPINGS as _WAVE_CLASSES,
     NODE_DISPLAY_NAME_MAPPINGS as _WAVE_NAMES,
 )
+from .nodes.text import (
+    NODE_CLASS_MAPPINGS as _TEXT_CLASSES,
+    NODE_DISPLAY_NAME_MAPPINGS as _TEXT_NAMES,
+)
 
-NODE_CLASS_MAPPINGS = {**_JUDGE_CLASSES, **_RECEPTOR_CLASSES, **_AUDIO_CLASSES, **_WAVE_CLASSES}
+NODE_CLASS_MAPPINGS = {**_JUDGE_CLASSES, **_RECEPTOR_CLASSES, **_AUDIO_CLASSES,
-NODE_DISPLAY_NAME_MAPPINGS = {**_JUDGE_NAMES, **_RECEPTOR_NAMES, **_AUDIO_NAMES, **_WAVE_NAMES}
+                       **_WAVE_CLASSES, **_TEXT_CLASSES}
+NODE_DISPLAY_NAME_MAPPINGS = {**_JUDGE_NAMES, **_RECEPTOR_NAMES, **_AUDIO_NAMES,
+                             **_WAVE_NAMES, **_TEXT_NAMES}
 
 WEB_DIRECTORY = "./web"  # serves web/audio_wave.js (the waveform + segment widget)
 __all__ = ["NODE_CLASS_MAPPINGS", "NODE_DISPLAY_NAME_MAPPINGS", "WEB_DIRECTORY"]

nodes/text.py

@@ -0,0 +1,27 @@
+"""
+Text (Prompt) node — a plain multiline STRING source.
+
+Needed because the judge's system_prompt / user_prompt / axes / reference_description
+are input sockets (forceInput): wire this node into them to type a value. Handy for the
+LTX system prompt, custom axes, etc.
+"""
+
+from __future__ import annotations
+
+
+class CalibratorText:
+    CATEGORY = "prompt_calibrator"
+    FUNCTION = "out"
+    RETURN_TYPES = ("STRING",)
+    RETURN_NAMES = ("text",)
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {"required": {"text": ("STRING", {"default": "", "multiline": True})}}
+
+    def out(self, text):
+        return (text,)
+
+
+NODE_CLASS_MAPPINGS = {"CalibratorText": CalibratorText}
+NODE_DISPLAY_NAME_MAPPINGS = {"CalibratorText": "Text (Prompt)"}

workflow/workflow_ltx_audio_api.json

@@ -0,0 +1,71 @@
+{
+  "1": {
+    "class_type": "LoadImage",
+    "inputs": {
+      "image": "reference.png"
+    },
+    "_meta": {
+      "title": "Reference subject image (put in ComfyUI/input/)"
+    }
+  },
+  "2": {
+    "class_type": "AudioWaveSegments",
+    "inputs": {
+      "audio": "audio.wav",
+      "fps": 24,
+      "max_segments": 6,
+      "notes": "",
+      "segments_json": "[]"
+    },
+    "_meta": {
+      "title": "Audio Wave + Segments (upload + click segments)"
+    }
+  },
+  "3": {
+    "class_type": "CalibratorText",
+    "inputs": {
+      "text": "You are an expert prompt engineer for the LTX-2.3 video model, generating a\nmulti-segment \"Prompt Relay\" timeline TIMED TO AUDIO. Output ONLY valid JSON — no\nmarkdown, no code fences, no text before or after.\n\nYou are given:\n- IMAGE 1 = the REFERENCE subject/scene. Use it ONLY for persistent visual appearance\n  (subject's look, setting, lighting, color palette, style) in global_prompt.\n- IMAGE 2 = the AUDIO ENERGY WAVEFORM (energy envelope with segment boundaries and beat\n  markers). Read the overall energy shape and where it peaks.\n- An \"AUDIO GUIDE\" text block: per-segment start, duration_s, frames (already snapped to\n  8n+1), energy level, stage hint, per-segment NOTE, and GLOBAL NOTES.\n\nMatch this schema exactly:\n\n{\n  \"fps\": 24,\n  \"global_prompt\": \"string\",\n  \"negative_prompt\": \"string\",\n  \"beats\": [\n    {\n      \"segment\": 1,\n      \"duration_s\": 5,\n      \"frames\": 121,\n      \"camera\": \"string\",\n      \"delta\": \"string\",\n      \"sound\": \"string\",\n      \"dialogue\": \"string\",\n      \"prompt\": \"string\"\n    }\n  ]\n}\n\nRules:\n- fps: take from the AUDIO GUIDE.\n- global_prompt: one present-tense paragraph with ALL persistent elements taken from\n  IMAGE 1 — subject's physical appearance (age, hair, clothing as visual cues), setting,\n  lighting, color palette, style — plus the sustained ambient/audio layer. No camera\n  moves. Never repeated in beats.\n- negative_prompt: \"shaky, glitchy, low quality, worst quality, deformed, distorted,\n  disfigured, motion smear, motion artifacts, fused fingers, bad anatomy, ugly, static\".\n- beats: EXACTLY ONE beat per AUDIO GUIDE segment, in order. Copy that segment's\n  \"segment\", \"duration_s\", and \"frames\" verbatim (do not recompute).\n- Beat 1: \"delta\" = the static visible starting state only; \"camera\" = \"static hold\".\n- Every later beat: \"delta\" = ONLY what CHANGES from the previous beat. Never repeat the\n  global_prompt or earlier beats.\n- MOTION FROM AUDIO: drive each beat's motion from its energy — higher energy => faster\n  motion and bigger camera moves; lower energy => slower, settle. Follow the energy\n  envelope across the timeline (establish -> build -> peak -> settle), putting the release\n  on the final segment.\n- NOTES OVERRIDE: if a segment has a NOTE, that instruction OVERRIDES the energy default\n  for that beat (e.g. NOTE \"slow\" on a HIGH-energy segment => slow, deliberate motion).\n  Apply GLOBAL NOTES throughout.\n- \"camera\" starts with a camera verb (push in, pan, arc, handheld drift, hold).\n- \"sound\" = short sound cue; \"dialogue\" = spoken line in quotes or \"\".\n- \"prompt\" = the final assembled beat text in order: camera, then delta action, then any\n  dialogue/sound — present tense, concrete on-screen verbs, no emotion labels. Ready to paste.\n"
+    },
+    "_meta": {
+      "title": "LTX audio system prompt"
+    }
+  },
+  "4": {
+    "class_type": "QwenVLImageJudge",
+    "inputs": {
+      "reference_image": [
+        "1",
+        0
+      ],
+      "mode": "chat",
+      "profile": "general",
+      "model_select": "(manual — use model_path below)",
+      "model_path": "3.5-9b",
+      "precision": "bf16",
+      "max_new_tokens": 3072,
+      "temperature": 0.4,
+      "swap_eval": false,
+      "enable_thinking": false,
+      "json_output": true,
+      "keep_loaded": true,
+      "auto_download": true,
+      "report_dir": "/media/p5/Comfyui/output/ltx",
+      "run_tag": "ltx",
+      "generated_image": [
+        "2",
+        0
+      ],
+      "system_prompt": [
+        "3",
+        0
+      ],
+      "user_prompt": [
+        "2",
+        1
+      ]
+    },
+    "_meta": {
+      "title": "Judge (chat + json_output) -> LTX beats JSON"
+    }
+  }
+}