OmniHuman 1.5 is an audio-driven video model for talking-head content: starting with a single portrait photo and a voice clip, it makes a static image speak while coordinating lip movements, expressions, and head-and-shoulder motion. It is suitable for product explainers, course introductions, and brand announcements. You can adjust emotion and style with prompts, and use subject masks to specify the on-camera person in group photos.
Submit requests to the public API at api.acedata.cloud using the documented parameters, then use the results in your application.
api.acedata.cloudomnihuman-1.5Input parameters and result formats vary by service. Use the public API for this model and follow its guide for generation, task retrieval and editing operations.
The above are this platform's asset requirements and invocation methods; the recommended audio duration does not represent the model's native limit.
OmniHuman 1.5 does more than animate the mouth in a photo: it coordinates expressions and head-and-shoulder movements with the rhythm of the voice. The creative focus is on how the person delivers the content, making it suitable for talking-head videos that require a consistent presenter; when preparing audio, first determine pauses, emphasis, and the overall delivery tone.
In addition to the photo and voice, you can use prompts to describe emotions such as gentleness, calmness, and naturalness, and request subtle head movements or stable performance. This guides the person's performance style, making it suitable for creating course explanations, brand announcements, and warm content separately, rather than replacing the dialogue in the audio.
Use a clear front-facing portrait as the visual foundation, without preparing multi-angle modeling assets in advance. Group photos can be paired with subject masks to specify the driven subject; after generation, receive a video URL to continue downloading, editing, or adding it to product pages, allowing talking-head content to fit into existing content production workflows.
Turn product updates into voiceovers, pair them with a fixed presenter headshot, and generate talking-head clips introducing new features. Then combine them with screen recordings, subtitles, and key annotations to deliver videos for help centers or new-user onboarding, with the person handling narration and the screen recording demonstrating actual operations.
Record course openings, knowledge-point overviews, or training reminders as audio, pair them with an instructor photo, and generate chapter-based digital human explanations. Using the same avatar maintains visual continuity across the course; when formulas, charts, or steps are involved, add courseware visuals in post-production instead of making the person present all the information.
Prepare different audio for event announcements, product introductions, or frequently asked questions, and use the same branded persona to generate separate videos. Deliverables can continue to include product footage, subtitles, and end cards; reusing the photo can reduce repeated filming, but each piece of content should still be checked to ensure expressions, lip movements, and audio match.
If you already have a usable portrait photo and finalized voiceover, and the goal is to have the person speak directly to the audience, OmniHuman 1.5's single-image audio workflow is straightforward. It is better suited to performances organized around sound rather than building complex narratives from text. Choose 1.5 based on talking-head needs and sample results; there is no need to interpret the version number as a comprehensive improvement for every task.
For presenter explanations, course hosting, and brand announcements, prioritize OmniHuman 1.5; if the focus is on large movements, character positioning, scene changes, or precisely recreating an action sequence, consider animation or video models that support the relevant controls. For talking-head tasks, first test photos, voices, and prompts with short clips before producing the full content.
Choose a clear portrait, prepare MP3/WAV audio, and verify the script. Both the image and audio must be publicly accessible; it is recommended to test with a short talking-head segment first.
Provide model=omnihuman-1.5, image_url, and audio_url to /dreamina/videos; prompt can describe emotion and stability, and for multi-person scenes provide mask_url according to the documentation.
Save the task_id asynchronously, query /dreamina/tasks or receive a callback; retrieve video_url, compare it against the audio to check mouth movements, pauses, and head-and-shoulder motion, then add subtitles.
Use a clear single-person portrait and a pre-recorded course introduction voiceover. The person speaks naturally, with a calm expression and slight head and shoulder movement, while the background remains stable.
Listen to the audio first, then check lip sync, pauses, and expressions; both image and audio URLs must be provided, and the prompt cannot replace the audio.
This workflow requires a portrait image and driving audio; text alone cannot replace these two assets. You can first record or create the script as an mp3/wav, then submit the audio URL. The prompt is used to adjust expressions, emotions, and style; it does not directly read the lines aloud.
It is recommended to use a clear, front-facing portrait with good lighting and no obstructions. The person should occupy an appropriate portion of the image. The photo must have a publicly accessible URL. Before formal production, you can test the photo with a short audio clip to check lip movements, expressions, and head-and-shoulder motion, then continue using the asset.
You can submit an array of subject mask URLs through mask_url to specify the person to drive in a multi-person photo; even if there is only one mask URL, it should still be submitted as an array. You should first identify the target and prepare the corresponding mask. This feature is for subject selection and should not be understood as allowing multiple characters to speak separately or complete a multi-person dialogue in a single request.
The voice itself provides speaking speed, pauses, and delivery rhythm, while the prompt can add requirements such as a gentle, calm, natural manner or slight head movements. It is recommended to keep the voice-over and prompt consistent, avoiding an excited voice with text requesting calmness; the final performance should still be checked through the generated video.
Submit image_url and audio_url to POST /dreamina/videos, and specify omnihuman-1.5. Results can be retrieved synchronously for short tasks; for longer tasks, you can use callback_url, or set async:true and then query /dreamina/tasks. After completion, read video_url and save it.
Choose a development approach or application suited to this model and follow its setup guide.
Python SDK Integration Guide
SDK installation, authentication and examplesTypeScript SDK Integration Guide
SDK installation, authentication and examples