Drama 3 Editor:
Drama3Editor turns a generated Drama 3 take into an editable audio timeline so you can refine its pacing, remove unwanted sections, and balance its voices without generating the performance again. Its editor opens over ComfyUI and presents one track per voice, with captions that highlight either whole lines or individual words when word alignment is available. You can adjust silence handling, review flagged lines, make cuts, add silence, name voices, and change their volume. Rather than simply deleting an entire long pause, its pause-shortening method removes the middle, preserving the end of the preceding sound and the beginning of the next. Gaps between dialogue lines have separate controls, allowing closely packed lines to gain breathing room while overly long exchanges become tighter. You can also trim the beginning and end and optionally normalize the final peak level. Choosing “Use these edits” runs the node again with your changes, and those changes remain stored in the workflow.
Drama 3 Editor Input Parameters:
codec
A required DRAMA3_CODEC connection that supplies the codec used to turn the take’s encoded audio into a waveform the editor can analyze and edit. This is the decoding component, not the voice or dialogue itself. It must be compatible with the supplied codes; otherwise, decoding can fail before the timeline or edited audio is produced. Connect a compatible codec from the Drama 3 codec-loading workflow. The interface defines no default value, minimum, maximum, or selectable options for this connection.
codes
A required DRAMA3_CODES connection containing the encoded take and its dialogue-line information. The node decodes the audio and reads the take’s lines information to analyze the recording and organize its timeline. Consequently, this connection determines both the underlying performance you hear and the dialogue structure available for editing; it is not a replacement for an ordinary audio-file input. Connect compatible codes produced by the Drama 3 generation or code-loading workflow. The interface defines no default value, minimum, maximum, or selectable options for this connection.
aligner
An optional QWEN3_ASR_MODEL connection for the Qwen3 forced aligner supplied through Load Qwen3-ASR. It provides word-level timing so captions can light up as individual words are spoken, making it easier to inspect where speech occurs when editing a take. Without it, the editor still works, but caption highlighting follows whole lines instead of individual words. This connection changes the timing detail available during analysis; it does not regenerate or replace the spoken performance. Its execution default is None, meaning no aligner is supplied, and the interface defines no numeric limits or selectable options.
edits
An optional STRING value that stores the editor’s settings and manual changes as JSON. It is hidden and socketless, so you normally manage it through the editor rather than through a visible text field or cable. The interface default is an empty string, which applies the built-in editing defaults; a nonempty value is parsed as JSON and merged over those defaults. The defaults are threshold_db = -50.0 for silence detection, trim_start = true with lead = 0.05 seconds, trim_end = true with tail = 0.15 seconds, pause_longer = 0.75 seconds and pause_to = 0.75 seconds for long-pause handling, and gap_min = 0.3 seconds and gap_max = 1.0 second for gaps between lines. Manual-edit collections keep, cuts, and inserts begin empty, speakers begins as an empty object, and normalize defaults to false. These are settings inside edits, not additional node inputs. When normalization is enabled, the rendered audio is scaled to a peak of -1 dBFS if its peak is nonzero; final samples are clamped to the range -1 to 1 regardless of that setting. The interface specifies no string-length limit, numeric ranges for the stored settings, or dropdown options. Because the supplied code does not document the full structure of manual-edit entries, use the editor to create them rather than guessing their JSON format.
Drama 3 Editor Output Parameters:
AUDIO
The single AUDIO output contains the decoded take after the current editing settings have been applied, including any planned trimming, pause adjustments, cuts, inserted silence, and supported rendering changes. It is a ComfyUI audio object with a waveform tensor and a sample_rate value, suitable for compatible downstream audio playback, saving, or processing nodes. The node packages the waveform with one batch dimension and one channel dimension, producing a single mono audio item. Its sample rate is the extension’s SAMPLE_RATE constant; the supplied source does not show its numeric value. The resulting duration can differ from the original take because removing audio shortens it and inserting silence lengthens it. The editor preview is delivered separately as interface information and is not an additional output socket.
Drama 3 Editor Usage Tips:
- Start with a compatible codec and a complete generated or loaded take. Drama3Editor decodes and edits existing codes; it does not replace the generation stage or accept ordinary audio directly.
- Run the node to obtain its editor preview, make your adjustments, and choose “Use these edits” to apply them through another execution. Save your workflow afterward so the stored editing settings remain available.
- Connect the optional aligner when you need word-by-word caption highlighting for detailed inspection. Leave it disconnected when line-level highlighting is sufficient; it is not required for editing.
- Listen to the default result before changing timing settings. An empty
editsstring still enables start and end trimming, long-pause handling, and inter-line gap adjustment, so it does not necessarily produce an untouched decode. - For tighter pacing, use the editor’s long-pause controls to lower the target pause duration relative to the threshold for identifying long pauses. For example, a threshold of 0.75 seconds and a target of 0.4 seconds expresses the intention to shorten pauses longer than 0.75 seconds to 0.4 seconds; these are example settings, not documented interface limits.
- Adjust gaps between lines separately from pauses within a line. The default inter-line range is 0.3–1.0 seconds; use this distinction to give speaker changes enough room without making every pause in a performance equally long.
- If quiet speech or soft word endings are being treated as silence, try a more negative
threshold_dbvalue and review the result. If too much low-level background sound remains classified as audible content, try a less negative value cautiously. - Check the start and end after trimming. The default retained margins are 0.05 seconds at the beginning and 0.15 seconds at the end; increase the relevant margin or disable that trim if the performance needs more breathing room or a longer decay.
- Use the editor’s keep controls for intentional pauses, and audition transitions after manual cuts. Although automatic pause shortening preserves the edges of a pause, dramatic timing still benefits from listening rather than relying only on the waveform display.
- Enable normalization when you want a consistent final peak level of -1 dBFS. Peak normalization adjusts overall amplitude; it does not guarantee equal perceived loudness between voices, so use the voice-volume controls when one speaker needs balancing against another.
Drama 3 Editor Common Errors and Solutions:
Required input missing: codec or codes
- Explanation: Both connections are mandatory. ComfyUI may report a missing-input validation error before the node executes; the exact message depends on your ComfyUI version.
- Solution: Connect a compatible
DRAMA3_CODECsource tocodecand a validDRAMA3_CODESsource tocodes, then run the workflow again.
JSONDecodeError: Expecting property name enclosed in double quotes
- Explanation: A nonempty
editsvalue is processed with Python’s JSON parser. Hand-edited settings containing single-quoted keys, trailing commas, or other invalid JSON syntax can stop execution. This is a possible parser message, not a custom Drama3Editor error. - Solution: Prefer making changes through the editor. If you intentionally edit the stored JSON, use double-quoted property names and valid JSON syntax. Resetting
editsto an empty string restores the built-in defaults but discards the stored edits.
JSONDecodeError: Expecting value
- Explanation: The stored
editstext is nonempty but does not contain a valid JSON value, or one of its settings has a missing value. Whitespace-only text can also trigger this error because it is passed to the JSON parser. - Solution: Restore valid editor-generated settings, or use a genuinely empty string to reset the settings. When editing JSON manually, use JSON values such as
trueandfalserather than Python-styleTrueandFalse.
KeyError: 'lines'
- Explanation: The node explicitly reads
codes["lines"]during analysis. If the supplied codes object does not include that field, execution can fail even if it contains some encoded audio. - Solution: Use a complete
DRAMA3_CODESobject from a compatible Drama 3 generation or loading workflow. If the object was modified or converted outside that workflow, regenerate or reload it with its dialogue-line information intact.
Decoding fails before the editor opens
- Explanation: Decoding is the first processing step. An incompatible codec, damaged codes, or a problem in the upstream loading stage can therefore prevent both analysis and preview creation. The supplied source does not define a specific decoding error message.
- Solution: Check that both upstream nodes finish successfully and that the codec is compatible with the take. Try decoding the same take with Drama3Decode to help determine whether the problem occurs before editing.
Captions highlight whole lines instead of individual words
- Explanation: This is expected behavior when no aligner is connected, not necessarily an error. The interface explicitly supports whole-line highlighting as the fallback.
- Solution: Connect the Qwen3 forced aligner through the optional
alignerinput and rerun the node if you need word-level highlighting.
Quiet speech is shortened or pauses sound unnatural
- Explanation: Silence detection and timing settings may not suit the recording. Quiet speech can be mistaken for silence, while aggressive pause targets or inter-line gap settings can change the intended delivery.
- Solution: Try a more negative silence threshold, preserve intentional pauses with the editor’s keep controls, and relax the pause or gap settings. Review the affected lines after applying the changes.
The beginning or end sounds clipped
- Explanation: Start and end trimming are enabled by default, with short retained margins. A soft onset, breath, or lingering decay may need more space than those defaults provide.
- Solution: Increase the beginning or ending margin in the editor, or disable the relevant trim option, then rerun and audition the boundary.
The edited audio sounds louder than expected
- Explanation: Optional normalization scales any nonzero result to a peak of -1 dBFS, which can substantially raise a quiet take. Voice-volume adjustments can also alter the balance before the final result is clamped.
- Solution: Disable normalization if you want to retain the rendered level without peak scaling, and check the voice-volume settings. Use normalization for overall peak control rather than as a substitute for balancing individual voices.
