Seed Audio 1.0: Text + Image to Audio
Upload a character image and write a text prompt to generate audio. The model derives the voice from your image, then produces speech, ambience, background music, and sound effects.
About
This workflow turns a single character image and a text prompt into a fully mixed audio clip with speech, ambience, background music, and sound effects. It centers on the ByteDanceSeedAudio node (Seed Audio 1.0), which accepts your image from LoadImage as a visual reference to derive a matching vocal identity, then synthesizes the spoken script and supporting layers described in your prompt. The output is saved with SaveAudioAdvanced, so you leave ComfyUI with a ready-to-share audio file.
What you provide
- Reading the workflow
Runs in ComfyUI. This workflow can't run on this page. Open it in Comfy Cloud with everything loaded, or download it for your own ComfyUI.
Try in CloudQuestions & answers
The image conditions the model to infer a fitting vocal identity (e.g., timbre and style cues) for the character. Use a clean, single-subject portrait with good lighting and minimal occlusion for more stable, consistent results.
Control them via the text prompt—be explicit about mood, instruments, intensity, and placement (e.g., soft piano underscore, distant thunder, quiet cafe ambience). The model outputs a single mixed track; stems aren’t provided by this workflow.
Keep the same image and prompt, and if the node exposes a seed parameter, fix it to reproduce a take. To explore alternatives, change the seed or make small prompt edits (pace, emotion, music style) and re-run.
It depends on your ByteDanceSeedAudio node build. Some variants require API credentials or additional model setup. Open the node, check its fields/tooltips, and follow the extension’s installation notes before running the graph.