A recognizable speaker
Keep a clean, permitted voice reference and a stable character name. Check the same line in two different emotional deliveries before committing to a longer script.
AUDIO MODEL GUIDE · SEPTEMBER 2026
Seed Audio 2.0
Know the version. Shape your next scene.
Start with the sound you need. Understand the published capabilities, choose useful references, and turn an idea into a scene brief you can actually review.
1.0 on SeedAudioText, audio references or a still image
A clearer briefWords, voices and sound cues
Version-awareProvider descriptions kept separate
A note on the 2.0 name
Checked September 24, 2026: Dreamina's 2.0 model address redirects to Seedaudio 1.5, marked coming soon. Some of its audio feature pages still use the 2.0 name. We cannot confirm a generally available 2.0 release from these conflicting pages. SeedAudio's current model integration is 1.0.
See the source notes01 / SCENE BRIEF
Make one small scene worth hearing.
Use these original writing exercises to decide what you want to hear. Edit a short passage, separate each speaker's direction, and keep a single sound cue easy to judge. These are prompts, not generated recordings.
Your scene brief
Prompt Guide290/2048
A starter replaces only the text. Copy your finished brief, then sign in to open the homepage editor and paste it there. Check your voice and reference settings before generating; your text is not transferred automatically.
Voice, references & output
Voice direction
Describe a voice in the brief, or choose a preset in the main editor. Keep the speaker mapping consistent.
Reference material
The existing editor has audio reference controls and a still-image alternative. It does not accept a video timeline.
Output choices
Format, sample rate, speed, volume and pitch stay in the shared editor. Choose only what your next editing step needs.
The existing SeedAudio editor uses the 1.0 workflow. This guide does not add video uploads, six-minute generation or separate stems. A duration in a prompt is a creative target, not a guaranteed result.
02 / CREATIVE CONTROL
Choose the control that solves your problem.
A bigger version number is not an editing plan. Write down what makes the current take unusable, then ask which input or output control would let you fix it without starting over.
Sound that meets the cut
Mark the action you need to hear: a footstep at the doorway, a reply after a glance, a latch at the end. A still image gives context; it does not supply the timing of a moving edit.
A mix you can revise
Decide whether you need a finished combined file or independent tracks. If music covers a word, a new full take and a separate dialogue track are very different editing options.
03 / INPUT MODES
Start from the material you have.
Dreamina describes four input combinations. The labels below explain the workflow, not a promise that every option is available on SeedAudio or in your provider account.
Text → audio
Start with the exact words, who says them and one audible event. Useful when exploring a fictional voice or a scene that has not been recorded.
Prepare: script + sound directionText + audio → audio
Use a clean voice reference when identity matters. Label the speaker and describe the delivery you want, keeping background noise out of the reference.
Prepare: permitted recording + scriptText + video → audio
A video-conditioned workflow can use a moving scene as context. Keep the cut stable and list the visible events that need sound. SeedAudio has no video input.
Prepare: picture edit + cue listText + audio + video
For a scene that needs both a known cast and a picture edit, keep each speaker's recording separate and map names consistently. Verify this mode with the provider.
Prepare: cast references + final cut04 / MODEL & WORKSPACE
Separate the model description from your tool.
The Dreamina column summarizes its current 1.5 model page. Related feature pages still use 2.0. Published limits do not establish availability, account access or measured output quality.
Area
SeedAudio · 1.0 workflow
Dreamina · published description
Input
SeedAudio · 1.0 workflowText with audio references, or a still image; no video control.
Dreamina · published descriptionFour input modes, including video context.
References
SeedAudio · 1.0 workflowUp to three audio references in the existing editor; audio and image are alternatives.
Dreamina · published descriptionUp to six reference recordings.
Length
SeedAudio · 1.0 workflowThe official 1.0 model describes a two-minute ceiling. Actual generation depends on the service and account.
Dreamina · published descriptionUp to six minutes in its model description.
Editing
SeedAudio · 1.0 workflowAn audio-file result; no separate-stem export control in the current workflow.
Dreamina · published descriptionTimestamp controls and separate tracks described.
Access
SeedAudio · 1.0 workflowExisting 1.0 editor and API documentation; no 2.0 model selector.
Dreamina · published descriptionModel page currently says coming soon; check actual account access.
05 / FROM WORDS TO SOUND
Write the scene in the order we hear it.
Keep the first test small enough that you can tell what changed. This original observatory exercise is a writing example, not a benchmark or a recording made with Seed Audio 2.0.
Say exactly what is spoken
Put dialogue in quotation marks. Keep one line per speaker and remove stage directions that might be read aloud. Decide which words must remain unchanged.
Give each voice a role
Name the speaker, describe the voice once, and attach emotion to the line. A calm character can still sound surprised. Keep casting and acting instructions distinct.
Place the sound cue
Specify whether the relay clicks before the first line, between the speakers or after the reply. Keep the room tone underneath rather than competing with the words.
Define a clean finish
Choose what should remain after the last word. For a clip that will be edited, listen for a clipped ending, unwanted extra dialogue or a background that stops abruptly.
ORIGINAL EXERCISE
Observatory at dawn
Room tone
Iris speaks
Ren replies
Relay click
Quiet ending
English. A brief exchange in a small observatory at dawn. Iris has a calm, low voice: "The sky is clearing." Ren, bright and curious, replies: "Then let's begin." A soft relay clicks after the reply. Keep the voices close and distinct, with a faint ventilation bed. No music or extra words.
06 / LISTEN & REFINE
One clear change. One more listen.
Save each brief beside its result. Compare the same passage at a comfortable volume and decide whether the change helped before adding another voice, sound or direction.
The words feel rushed
Reduce the number of lines. Read the script aloud once to check its natural pace. Asking for a longer clip in the text cannot override the service's output limit.
The speakers blend together
Check names and reference assignments. Make the delivery directions distinct, then listen to each voice on its own. Do not hide a weak voice match under music.
The background masks speech
Remove the music instruction and keep one quiet ambience. Ask for less density before adding volume. A mixed file does not become separate tracks when you download it.
07 / QUESTIONS
Before your next audio project.
No. SeedAudio's existing integration is Seed Audio 1.0. This page explains version naming and scene preparation; it does not enable a new model.
At our September 24, 2026 check, Dreamina's 2.0 model URL redirected to a 1.5 page, while related audio pages still said 2.0. The naming is inconsistent. Check the actual product and account before relying on a version-specific feature.
Those controls are not offered in the current SeedAudio workflow. The Dreamina descriptions are external capabilities, not additions to this editor. A still image is not a substitute for a video timeline.
No. A written duration is a request, not an entitlement or a reliable length control. The six-minute statement belongs to Dreamina's description. Check the selected service and account before planning a long scene.
No. The starters are original writing exercises. The player contains an existing 1.0 orbital escape sample with its own full prompt. Changing the exercise never changes that recording.
The current SeedAudio API documentation describes the 1.0 integration. Changing a name in a request does not add a model. Use the documented fields and check the actual model before building a video workflow.
Use a recording you have permission to use, keep a record of that permission, and review the result for unwanted imitation. Check the applicable provider terms before publishing. This guide does not grant rights to a person's voice.
08 / ORIGINAL SOURCES
Read the source. Check the product.
Reviewed September 24, 2026. Version claims come from the links below. The writing exercises, listening advice and conceptual artwork are editorial material, not proof of model performance.
01020304
Dreamina model page
Now titled Seedaudio 1.5; the previous 2.0 URL redirects here.
Dreamina audio overview
Still uses the 2.0 name. Read alongside the model page.
ByteDance Seed · 1.0
Original model announcement and technical context.
fal · Seed Audio 1.0 API
Provider fields for the current model integration.
Start with one scene you can judge.
Write a clear brief, keep the version in view, and make the next take answer one useful question.