Eve library

ISIS v2 Voice Line Pack Image-Conditioned Workflow Graph

Guide

Revision 1

On this page

Generated: 2026-03-03T11:56:56.509Z Checklist: MXA-0543 (ISIS2-00741) Asset Family: voice-line-pack Phase: Phase 4 | Milestone: M3 Launch (2026-07-27 to 2026-10-30)

Objective#

Define an image-conditioned voice-line-pack workflow graph that maps image-conditioned speaker/style references to provider-capable conditioning guidance while preserving deterministic synthesis defaults and mastered-dialogue delivery contracts.

Deterministic Defaults#

Setting Value
modelName xtts_v2_voiceset_a.safetensors
promptTemplateId voice-line-pack-image-conditioned-v1
styleProfileId voice-line-pack-style-anchor-v1
seed 2026073103
durationSeconds 8
speakingRate 1
pitchShiftSemitones 0
conditioningStrength 0.74
conditioningMode reference-image
providerCapabilityMode required
referenceMixStrategy weighted-blend
maxReferenceCount 4
preservePromptIntent true
voiceStability 0.74
sampleRateHz 48000
targetLufs -18
outputFormat wav
outputChannels 1
outputBitDepth 24

Input Contract#

Input Type Required Default
cue_id string yes "VOICE_LINE_M3_IMAGE_CONDITIONED_01"
script_text string yes "Commander, reroute auxiliary power to deck three and hold for extraction."
speaker_id enum yes "narrator_f"
language_locale enum yes "en-US"
emotion enum yes "confident"
speaking_rate float yes 1
pitch_shift_semitones float yes 0
duration_seconds integer yes 8
text_prompt_hint string no "Preserve confident delivery, clear diction, and measured cadence from references."
reference_image_url string yes "asset://references/images/voice-line-speaker-anchor-001.png"
reference_image_urls array no ["asset://references/images/voice-line-expression-anchor-002.png","asset://references/images/voice-line-lighting-anchor-003.png"]
speaker_style_reference_id string no "VOICE_LINE_STYLE_ANCHOR_V1"
conditioning_strength float yes 0.74
conditioning_mode enum yes "reference-image"
provider_capability_mode enum yes "required"
reference_mix_strategy enum yes "weighted-blend"
preserve_prompt_intent boolean yes true
prompt_template_id string yes "voice-line-pack-image-conditioned-v1"
style_profile_id string yes "voice-line-pack-style-anchor-v1"
sample_rate_hz integer yes 48000
target_lufs float yes -18
output_format enum yes "wav"
model_name string yes "xtts_v2_voiceset_a.safetensors"
voice_stability float yes 0.74
seed integer yes 2026073103

Graph Nodes#

Node ID Class Role Output Count
n1 LoadImage primary-reference-load 1
n2 LoadImage secondary-reference-load 1
n3 VoiceLinePromptComposer prompt-construction 1
n4 VoiceLineModelLoader model-load 1
n5 ImageVoiceConditioner reference-conditioning 2
n6 VoiceLineSynthesisNode audio-generation 1
n7 VoiceLineCleanupDenoise audio-post 1
n8 AudioLoudnessNormalize audio-post 1
n9 SaveAudio artifact-output 0

Graph Edges#

Edge ID From To
e1 n1:0 n5.primary_reference
e2 n2:0 n5.secondary_references
e3 n3:0 n5.prompt
e4 n5:0 n6.conditioned_prompt
e5 n5:1 n6.reference_conditioning
e6 n4:0 n6.model
e7 n6:0 n7.audio
e8 n7:0 n8.audio
e9 n8:0 n9.audio

Topological Node Order#

n1 -> n2 -> n3 -> n4 -> n5 -> n6 -> n7 -> n8 -> n9

Research Alignment#

Workflow Path Reference Conditioning Support Conditioning Input Keys Node Classes
SFX Sound Designer libs/isis/workflows/src/workflows/audio/sfx-sound-designer.json no conditioning_strength, conditioning_mode, provider_capability_mode AudioLoudnessNormalize, SaveAudio, StableAudioNode, StableAudioOpenModelLoader, StableAudioPromptComposer
Music Cue Composer libs/isis/workflows/src/workflows/audio/music-cue-composer.json no conditioning_strength, conditioning_mode, provider_capability_mode ACEModelLoader, ACEStepAudioGenerator, ACEStepPromptBuilder, AudioLimiter, SaveAudio
Voice Performance Synthesis libs/isis/workflows/src/workflows/audio/voice-performance-synthesis.json no conditioning_strength, conditioning_mode, provider_capability_mode AudioDeesser, KokoroModelLoader, KokoroPromptBuilder, KokoroTextToSpeech, SaveAudio

Acceptance Criteria#

  • Graph is acyclic and topologically sortable.
  • Reference conditioning path is explicit and includes required primary reference image and optional multi-reference stack.
  • Deterministic defaults are explicitly locked (seed, model, prompt template, style profile, conditioning mode/provider mode, loudness, and export format).
  • Output contract aligns to MXA-0541 required output classes for voice-line-pack.

Traceability#

  • Checklist item: MXA-0543
  • Ticket key: ISIS2-00741
  • Source references:
    • docs/domains/isis/remediation-v2/ISIS_REMEDIATION_V2_TICKETS.json
    • docs/domains/isis/remediation-v2/ISIS_V2_VOICE_LINE_PACK_REQUIREMENTS_BRIEF.json
    • docs/domains/isis/remediation-v2/ISIS_V2_VOICE_LINE_PACK_BASELINE_WORKFLOW_GRAPH.json
    • docs/domains/isis/remediation-v2/ISIS_V2_WORKFLOW_AUTHORING_GUIDE.json
    • libs/isis/workflows/src/workflows/audio/sfx-sound-designer.json
    • libs/isis/workflows/src/workflows/audio/music-cue-composer.json
    • libs/isis/workflows/src/workflows/audio/voice-performance-synthesis.json
    • docs/domains/isis/remediation-v2/ISIS_V2_AUDIO_CONDITIONING_SUPPORT.md
    • docs/domains/isis/remediation-v2/ISIS_V2_CONTROL_TAXONOMY.json
    • docs/domains/isis/remediation-v2/ISIS_V2_OUTPUT_PACKAGE_TAXONOMY.json
    • ISIS_TODOS_2.md