diff --git a/docs/PROVIDERS.md b/docs/PROVIDERS.md
index ca25a4a..a5cea35 100644
--- a/docs/PROVIDERS.md
+++ b/docs/PROVIDERS.md
@@ -213,11 +213,14 @@ Google TTS offers 700+ voices across 50+ languages. Voice names follow the patte
| Type | Example | Quality | Cost |
|------|---------|---------|------|
+| **Chirp 3 HD** | `en-US-Chirp3-HD-Orus` | **Best (2024, most natural)** | **Mid — default** |
| Standard | `en-US-Standard-A` | Good | Cheapest |
| WaveNet | `en-US-WaveNet-D` | Very good | Mid |
| Neural2 | `en-US-Neural2-D` | Excellent | Mid |
| Studio | `en-US-Studio-O` | Professional | Highest |
-| Journey | `en-US-Journey-D` | Conversational | Mid |
+| Journey | `en-US-Journey-D` | Conversational (long-form) | Mid |
+
+**Recommended voices:** `en-US-Chirp3-HD-Orus` (male, rich/cinematic), `en-US-Chirp3-HD-Aoede` (female, warm). These are Google's newest tier — most natural-sounding, uses the v1beta1 endpoint automatically.
**Languages include:** English (US, UK, AU, IN), Spanish, French, German, Italian, Portuguese, Japanese, Korean, Chinese (Mandarin, Cantonese), Arabic, Hindi, Russian, Dutch, Polish, Turkish, Vietnamese, Thai, Indonesian, and 30+ more.
diff --git a/pipeline_defs/documentary-montage.yaml b/pipeline_defs/documentary-montage.yaml
index 611edc1..c10df44 100644
--- a/pipeline_defs/documentary-montage.yaml
+++ b/pipeline_defs/documentary-montage.yaml
@@ -55,13 +55,13 @@ stages:
- Tone register is ONE value from the fixed list
- Duration and shape are concrete
- Music plan is present (MANDATORY — silent only if user explicitly opted out)
- - End-tag plan is present (MANDATORY — one philosophical closing line, rendered as Remotion end-card, unless user explicitly opted out)
+ - End-tag plan is present (MANDATORY — one philosophical closing line, rendered as Remotion end-card, default mode "overlay" on final scenes, unless user explicitly opted out)
- Narration plan is present (narration itself is OPTIONAL — absence is fine if music + visuals + end-tag carry the register)
success_criteria:
- Schema-valid brief artifact
- thematic_question present in metadata
- music_plan present in metadata (source may be `none` ONLY with explicit user opt-out note)
- - end_tag_plan present in metadata (text, palette, duration — may be `null` ONLY with explicit user opt-out note)
+ - end_tag_plan present in metadata (text, palette, duration, mode — may be `null` ONLY with explicit user opt-out note)
- name: scene_plan
skill: pipelines/documentary-montage/scene-director
@@ -129,8 +129,8 @@ stages:
- No two adjacent cuts share subject AND scale
- Transition vocabulary is at most 4 distinct values
- Music config is present (MANDATORY — silent only if brief.metadata.music_plan.source=none with explicit opt-out note)
- - End-tag cut is present at the tail (MANDATORY — rendered separately via Remotion, concatenated after body, unless brief.metadata.end_tag_plan is null with explicit opt-out note)
- - total_duration_seconds matches sum of cut durations INCLUDING the end-tag hold
+ - "End-tag timing is present (MANDATORY — overlay mode: end_tag.offset_seconds in edit_decisions; concat mode: end-tag appended after body. Unless brief.metadata.end_tag_plan is null with explicit opt-out note)"
+ - total_duration_seconds matches sum of cut durations (body only — end-tag overlay does not extend duration; concat mode adds tag duration)
- Every cut has a reason
success_criteria:
- Schema-valid edit_decisions artifact
@@ -160,15 +160,16 @@ stages:
checkpoint_required: true
human_approval_default: false
review_focus:
- - Output duration matches planned within 1s (body + end-tag inclusive)
+ - Output duration matches planned within 1s (body duration for overlay mode; body + tag for concat mode)
- Resolution matches target_platform canvas
- Uniform LUT applied across the timeline
- Music is mixed in (MANDATORY — silent output only with explicit user opt-out recorded in brief)
- - End-tag MP4 rendered via Remotion and concatenated at the tail (MANDATORY — absence only with explicit opt-out)
- - First and last frames verified (last frame must be the end-tag card unless opted out)
+ - "End-tag rendered via Remotion (MANDATORY — absence only with explicit opt-out). Default mode: overlay (ProRes 4444 with alpha composited on final scenes). Fallback: concat (opaque card appended after body)."
+ - "Overlay mode: extract a frame from the overlay region and verify text is visible over footage, not over black. Concat mode: last frame must be the end-tag card."
- No silent fallback from a motion-led promise
success_criteria:
- Schema-valid render_report artifact
- Output file exists and passes ffprobe validation
- render_report.end_tag_rendered = true (or explicit opt-out)
+ - render_report.end_tag_mode = "overlay" or "concat"
- render_report.music_mixed = true (or explicit opt-out)
diff --git a/remotion-composer/src/Root.tsx b/remotion-composer/src/Root.tsx
index bbc163e..5d8985d 100644
--- a/remotion-composer/src/Root.tsx
+++ b/remotion-composer/src/Root.tsx
@@ -220,6 +220,24 @@ export const Root: React.FC = () => {
fadeOutSeconds: 0.6,
} as EndTagProps}
/>
+
>
);
};
diff --git a/remotion-composer/src/components/EndTag.tsx b/remotion-composer/src/components/EndTag.tsx
index 16447d5..6ffbfef 100644
--- a/remotion-composer/src/components/EndTag.tsx
+++ b/remotion-composer/src/components/EndTag.tsx
@@ -13,6 +13,12 @@ export interface EndTagProps {
fadeInSeconds?: number;
holdSeconds?: number;
fadeOutSeconds?: number;
+ // Overlay mode: render on a transparent background so the tag can be
+ // composited on top of the body footage in post, instead of being
+ // concatenated as a standalone black card. When `overlay=true` the
+ // AbsoluteFill drops its background fill — caller is responsible for
+ // rendering with an alpha-capable codec (VP9/WebM or ProRes 4444).
+ overlay?: boolean;
}
const PALETTES = {
@@ -49,6 +55,7 @@ export const EndTag: React.FC = ({
fadeInSeconds = 0.6,
holdSeconds = 4.3,
fadeOutSeconds = 0.6,
+ overlay = false,
}) => {
const frame = useCurrentFrame();
const { fps } = useVideoConfig();
@@ -113,7 +120,7 @@ export const EndTag: React.FC = ({
return (
/renders/body.mp4`.
-2. Render the end-tag via Remotion CLI with component-specific props:
- `npx remotion render EndTag --props='{"text":"...", "palette":"...","durationInFrames":132}' projects//renders/end_tag.mp4`
+ Save as `projects//renders/body.mp4`. Note the body fps.
+2. Compute `durationInFrames = round(duration_seconds × body_fps)`.
+3. Render the end-tag with alpha via Remotion CLI:
+ ```bash
+ npx remotion render src/index.tsx EndTagOverlay \
+ projects//renders/end_tag_overlay.mov \
+ --codec=prores --prores-profile=4444 \
+ --pixel-format=yuva444p10le --image-format=png \
+ --props='{"text":"...","palette":"...","overlay":true,
+ "fadeInSeconds":1.0,"holdSeconds":3.0,"fadeOutSeconds":1.5}'
+ ```
+ Use the `EndTagOverlay` composition with `overlay: true`. This
+ produces a ProRes 4444 MOV with a real alpha channel
+ (pix_fmt=yuva444p12le). Canvas must match body canvas.
+4. Compute the overlay offset:
+ - Read `edit_decisions.end_tag.offset_seconds` if present.
+ - Otherwise auto-compute: `offset = body_duration - tag_duration`.
+ The tag's fade-out should align with the body's closing fade-out.
+5. Composite via FFmpeg overlay with `-itsoffset`:
+ ```bash
+ ffmpeg -y \
+ -i body.mp4 \
+ -itsoffset {offset} -i end_tag_overlay.mov \
+ -filter_complex "[0:v][1:v]overlay=0:0:format=auto:eof_action=pass[v]" \
+ -map "[v]" -map "0:a" \
+ -c:v libx264 -preset medium -crf 18 -pix_fmt yuv420p \
+ -c:a aac -b:a 192k \
+ projects//renders/final.mp4
+ ```
+ `eof_action=pass` means the body video continues after the overlay
+ ends. The overlay's own alpha handles the fade-in/hold/fade-out.
+
+**Verification:** Extract a frame from the overlay region (e.g.
+`offset + 2s`) and confirm text is visible over footage, not over
+black. If the frame shows a black background behind the text, the
+alpha channel was lost — re-render with `--image-format=png`.
+
+#### Path B — Concat Mode
+
+Classic tail-card: opaque black card appended after the body. Use
+this only when `end_tag_plan.mode == "concat"`.
+
+**Execution:**
+
+1. Compose the body as above.
+2. Render the end-tag as opaque MP4:
+ ```bash
+ npx remotion render src/index.tsx EndTag \
+ projects//renders/end_tag.mp4 \
+ --props='{"text":"...","palette":"...","durationInFrames":132}'
+ ```
(5.5s at 24fps = 132 frames). Canvas must match body canvas.
-3. Concat body + end_tag with `ffmpeg -f concat -safe 0 -i list.txt -c copy final.mp4`
- or, if encoders don't match, re-encode with the documentary spec.
+3. Concat body + end_tag:
+ ```bash
+ ffmpeg -f concat -safe 0 -i list.txt -c copy final.mp4
+ ```
+ Or re-encode if codecs don't match.
+
+#### Common Rules (Both Modes)
**End-tag is MANDATORY.** The ONLY way to skip it is an explicit user
opt-out recorded as `end_tag_plan: null` with an `end_tag_opt_out_reason`.
@@ -179,7 +238,9 @@ a contract violation. Stop and surface before finalizing.
Record in `render_report`:
- `end_tag_rendered: true | false`
-- `end_tag_path: "projects//renders/end_tag.mp4"`
+- `end_tag_mode: "overlay" | "concat"`
+- `end_tag_path: "projects//renders/end_tag_overlay.mov"` (or `.mp4` for concat)
+- `end_tag_offset_seconds: ` (overlay mode only)
- `end_tag_text: "..."` (for audit trail)
If the brief says "no music" and the edit correctly has no music
diff --git a/skills/pipelines/documentary-montage/edit-director.md b/skills/pipelines/documentary-montage/edit-director.md
index f9f68f7..e26496e 100644
--- a/skills/pipelines/documentary-montage/edit-director.md
+++ b/skills/pipelines/documentary-montage/edit-director.md
@@ -225,6 +225,32 @@ by layering the outgoing clip's audio as an SFX entry in
Documentary montages with L-cuts feel 50% more coherent than ones
without. Use them on the 3-4 hardest transitions in the piece.
+### 8b. Place The End-Tag Overlay
+
+If `brief.metadata.end_tag_plan.mode == "overlay"` (the default), the
+end-tag will be composited on top of the final body footage at compose
+time. The edit director's job is to decide **when** the tag appears.
+
+Compute the offset: `offset_seconds = body_duration - tag_duration`.
+This makes the tag's fade-out align with the body's closing fade-out
+(the last cut's `transition_out: fade_out`). If the final cut's hold
+is shorter than the tag duration, start the tag earlier so it overlaps
+the second-to-last cut as well — this is fine and often looks better.
+
+Record in `edit_decisions.end_tag`:
+
+```json
+{
+ "end_tag": {
+ "offset_seconds": 84.5,
+ "notes": "Tag starts at body_duration - tag_duration. Aligns tag fade-out with final cut fade-out."
+ }
+}
+```
+
+If `mode == "concat"`, omit this section — the compose-director will
+append the tag after the body without needing a timing offset.
+
### 9. Emit The Edit Decisions
Canonical shape for this pipeline:
@@ -266,6 +292,10 @@ Canonical shape for this pipeline:
"ducking": false
}
},
+ "end_tag": {
+ "offset_seconds": 84.5,
+ "notes": "Tag starts at body_duration - tag_duration. Aligns tag fade-out with final cut fade-out."
+ },
"metadata": {
"pipeline": "documentary-montage",
"tone": "elegiac",
diff --git a/skills/pipelines/documentary-montage/idea-director.md b/skills/pipelines/documentary-montage/idea-director.md
index f854eb5..e6ef01a 100644
--- a/skills/pipelines/documentary-montage/idea-director.md
+++ b/skills/pipelines/documentary-montage/idea-director.md
@@ -95,7 +95,14 @@ defer this — it becomes an expensive surprise at the asset stage.
Every documentary-montage film closes on a philosophical end-tag — one
short, abstract line that gives the whole thing meaning. It is rendered
as a Remotion end-card ("shining underlined tag" register — bold weight,
-letter-spaced, animated underline) and concatenated after the last clip.
+letter-spaced, animated underline).
+
+**Default mode is `"overlay"`** — the tag fades in over the final scenes
+of the body footage, so it feels like part of the film rather than a
+separate card tacked on at the end. The alternative is `"concat"` which
+appends a standalone black-card after the body. Use concat only when the
+user explicitly asks for a separated title card, or when the final
+footage is too visually busy for legible text overlay.
**End-tag is MANDATORY.** The ONLY way out is an explicit user opt-out
recorded as `end_tag_plan: null` with an `end_tag_opt_out_reason` field.
@@ -110,12 +117,26 @@ Expected shape:
"palette": "warm_ivory_on_black",
"duration_seconds": 5.5,
"render_engine": "remotion",
- "component": "EndTag"
+ "component": "EndTag",
+ "mode": "overlay"
}
}
```
-Keep the copy to 3-9 words. It must be a thesis, not a summary.
+Fields:
+- `text` — 3-9 words. A thesis, not a summary.
+- `palette` — `"cool_offwhite_on_black"` or `"warm_ivory_on_black"`.
+- `duration_seconds` — total tag screen time (fade-in + hold + fade-out).
+ 5-8s is the sweet spot.
+- `render_engine` — always `"remotion"`.
+- `component` — always `"EndTag"`.
+- `mode` — `"overlay"` (default) or `"concat"`.
+ - **overlay**: tag rendered as ProRes 4444 with alpha → composited on
+ final body footage via FFmpeg overlay filter. Tag fades appear over
+ the last N seconds of live footage. The body's own fade-out and the
+ tag's fade-out should align.
+ - **concat**: tag rendered as opaque MP4 → appended after body via
+ FFmpeg concat. Total output duration = body + tag.
### 6. Note Narration Intent (OPTIONAL)
diff --git a/tools/audio/google_tts.py b/tools/audio/google_tts.py
index 09e0ec5..9e1fee2 100644
--- a/tools/audio/google_tts.py
+++ b/tools/audio/google_tts.py
@@ -77,8 +77,8 @@ class GoogleTTS(BaseTool):
"text": {"type": "string", "description": "Text to convert to speech"},
"voice": {
"type": "string",
- "default": "en-US-Neural2-D",
- "description": "Voice name (e.g. en-US-Neural2-D, en-US-Studio-O, en-GB-WaveNet-A)",
+ "default": "en-US-Chirp3-HD-Orus",
+ "description": "Voice name. Default tier is Chirp 3 HD (2024, most natural). Examples: en-US-Chirp3-HD-Orus (male, rich/cinematic), en-US-Chirp3-HD-Aoede (female, warm). Legacy tiers: en-US-Studio-O, en-US-Neural2-D, en-US-Journey-D.",
},
"language_code": {
"type": "string",
@@ -144,7 +144,7 @@ class GoogleTTS(BaseTool):
def estimate_cost(self, inputs: dict[str, Any]) -> float:
text = inputs.get("text", "")
char_count = len(text)
- voice = inputs.get("voice", "en-US-Neural2-D")
+ voice = inputs.get("voice", "en-US-Chirp3-HD-Orus")
# Pricing per million characters (approximate)
if "Chirp3-HD" in voice:
rate_per_char = 0.000030 # $30/1M chars
@@ -180,7 +180,7 @@ class GoogleTTS(BaseTool):
import requests
text = inputs["text"]
- voice_name = inputs.get("voice", "en-US-Neural2-D")
+ voice_name = inputs.get("voice", "en-US-Chirp3-HD-Orus")
language_code = inputs.get("language_code", "en-US")
speaking_rate = inputs.get("speaking_rate", 1.0)
pitch = inputs.get("pitch", 0.0)