Start one generation job without waiting; prop/character use the Asset Generation job lane, other supported kinds use the standalone lane. For audio, --mode selects sound_effect (default), music, or text_to_speech.
Dreamer project id (set only when Dreamer, the website agent, started the call)
--session-id <id>
Dreamer session id (set only when Dreamer, the website agent, started the call)
--loop
audio: request loopable output
--duration-seconds <n>
audio duration
--resolution <n>
material resolution: 1024 or 2048 (default "1024")
--repeat-per-meter <n>
material physical scale: tile repeats per metre
--splat-scope <scope>
Gaussian-splat scope: object or environment (default "environment")
--visibility <visibility>
character: public, unlisted, or private. Sent only when you pass it; otherwise the Asset Generation job driver applies its own default
--reference-review-seconds <n>
character: seconds the reference sheet waits for a look before it auto-approves (0-120). Unattended runs (an agent) should set it; without it the sheet waits for a person
--idempotency-key <key>
character: re-attach to the job a previous start created instead of creating a second (8-96 characters). Default: a fresh key, printed so a start whose answer was lost can be re-run safely
--max-lix <n>
refuse to start if the estimate exceeds this many LIX
--no-estimate
skip the up-front cost estimate
--mode <mode>
sound_effect, music, or text_to_speech (default "sound_effect")
--model <model>
an allowlisted ElevenLabs model id for the selected --mode
--output-format <format>
portable MP3 output format, e.g. mp3_44100_128
--prompt-influence <n>
sound_effect only: how closely to follow the prompt (0-1)
--force-instrumental
music only: request an instrumental-only composition
--text <text>
text_to_speech only: the text to convert to speech
--voice-id <id>
text_to_speech only: the voice id to speak with (see helix assets voices)
--language-code <code>
text_to_speech only: ISO 639-1 language code
--seed <n>
text_to_speech only: a seed for deterministic output
--stability <n>
text_to_speech only: voice stability (0-1)
--similarity-boost <n>
text_to_speech only: voice similarity boost (0-1)
--style <n>
text_to_speech only: voice style (0-1)
--speed <n>
text_to_speech only: voice speed (0.7-1.2)
--speaker-boost
text_to_speech only: enable speaker boost
--pronunciation-dictionary-ids <a,b,c>
text_to_speech only: up to 3 comma-separated pronunciation dictionary ids
Revise a job waiting at an approval gate, with new guidance. CHARGES a fresh regeneration — this is a spend, not a free retry. The only advance path when the server rejected its own concept.
Retry a FAILED job (no new guidance). Resumes from the stage that stopped — the mesh phase for a prop or character, the bundle/publish step for a material, image or texture
Poll one generation job without starting or advancing a duplicate
helix assets status [options] <job-id>
Argument
Required
Description
job-id
yes
Asset Generation job id
Option
Description
--pipeline <pipeline>
driver = a character driven by the Asset Generation job driver (start character), dreamer = prop/legacy character job lane (legacy name), asset = standalone lane (required)
Generate and publish a new prop or character with Asset Generation
helix assets generate [options] <kind> <prompt>
Argument
Required
Description
kind
yes
prop or character
prompt
yes
asset prompt
Option
Description
--title <title>
asset title
--additional-prompt <text>
additional generation guidance
--target-polycount <n>
target triangle count
--visibility <visibility>
public, unlisted, or private (default public for a prop; a character uses the Asset Generation job driver default unless you pass it)
--timeout <ms>
how long to wait for the job to finish before giving up (client-side only — the job keeps running; default 20min for a prop, 80min for a character)
--poll-interval <ms>
how often to poll job status while waiting (default 2000ms prop, 3000ms character)
--reference-review-seconds <n>
character: seconds the reference sheet waits for a look before it auto-approves (0-120). Unattended runs (an agent) should set it; without it the sheet waits for a person
--idempotency-key <key>
character: re-attach to the job a previous start created instead of creating a second (8-96 characters). Default: a fresh key, printed so a start whose answer was lost can be re-run safely
--max-lix <n>
refuse to start if the estimate exceeds this many LIX
Wait for an Asset Generation job and print its result. A character started with helix assets start character is driven by the server: this only reads it (progress lines on stderr, the job JSON on stdout; exit 0 for succeeded/held/handoff, 1 for failed or a client timeout). Older job ids (props, legacy characters) resume and publish through the previous loop
helix assets resume [options] <job-id>
Argument
Required
Description
job-id
yes
Asset Generation job id
Option
Description
--visibility <visibility>
public, unlisted, or private (legacy jobs: default is the visibility the job was generated with, else public). A driver job fixes its visibility when it starts: a different value here is an error
--timeout <ms>
how long to wait for the job to finish before giving up (client-side only — the job keeps running; default 80min for a driver job, 20min for a legacy job)
--poll-interval <ms>
how often to poll job status while waiting (default 3000ms driver job, 2000ms legacy job)
Decide at the open reference gate of an Asset Generation job driven by the server: approve it now, or revise it with guidance (a revise charges a regeneration)
Dreamer project id (set only when Dreamer, the website agent, started the call)
--session-id <id>
Dreamer session id (set only when Dreamer, the website agent, started the call)
--visibility <public|private>
where the image lands in the Vault. Default: your account preference (public unless you opted out). Use private for the concept/reference sheet of a private avatar, so the sheet is not published.
Generate a sound effect, music track, or spoken line via ElevenLabs (with Asset Generation), auto-publish it to Vault, and optionally save it locally. Blocks until done — use helix assets start audio for a pollable job.
helix assets generate-audio [options] [prompt]
Argument
Required
Description
prompt
no
generation prompt — required for sound_effect/music; text_to_speech uses --text instead
Option
Description
--mode <mode>
sound_effect, music, or text_to_speech (default "sound_effect")
--additional-prompt <text>
additional generation guidance
--model <model>
an allowlisted ElevenLabs model id for the selected --mode
--output-format <format>
portable MP3 output format, e.g. mp3_44100_128
--project-id <id>
Dreamer project id (set only when Dreamer, the website agent, started the call)
--session-id <id>
Dreamer session id (set only when Dreamer, the website agent, started the call)