An avatar’s performance prompt is not necessarily its speech input. If you placed the entire script in the acting description, inspect where the spoken audio actually comes from. Kling’s Avatar 2.0 guide separates a character image, speech audio and optional performance instructions. Uploading or generating speech is a distinct step from describing how the character behaves.
Trace the sentence to a real audio source
Find the approved line in the submitted materials. Is it in an uploaded recording, a text-to-speech input that produced an audio asset, or only the performance paragraph? Listen to the actual selected recording. A filename saying “new greeting” is insufficient if it still contains yesterday’s words. A text box beside the audio card may describe movement rather than create a soundtrack.
Keep script version and audio version together. If the visible text was corrected after audio generation, determine whether a new recording was actually made and selected. Changing text in a planning note does not prove that an existing waveform changed. Conversely, a correct audio file does not establish that the right file reached the avatar request.
Divide the brief by responsibility
For a hypothetical museum introduction, the approved speech might be “Welcome to the restored east gallery.” The performance direction could request a gentle smile and a restrained open-hand gesture. Put the sentence through the supported speech route and keep the acting note focused on observable behavior. Do not bury an alternative sentence in that note.
| Component | Authority | Inspection |
|---|---|---|
| Face and framing | Authorized source image | Correct person and usable composition |
| Words to speak | Approved recording or generated speech asset | Listen to complete line |
| Acting | Compatible performance description | Check expression and gesture |
| Combined result | Downloaded clip | Check timing and content together |
The example is a proposed preparation method. It does not report a generated museum clip or promise exact gestures.
Isolate the failure before another attempt
If the selected audio contains the wrong words, correct that source first. If the audio is right but the output contains different speech, preserve the request evidence and investigate the supported input path. If the speech is right but the performance is unsuitable, revise the acting direction without casually changing the approved recording.
Check the opening, pauses and ending, not only a short middle section. Look for a mouth moving while another sound plays, a gesture interrupting visibility, or the image being cropped too tightly for the intended action. Separate those visual defects from a wrong-script defect. A completed task state alone does not resolve any of them.
Hand over the three approved pieces
Save the source image, exact audio and final clip with matching version notes. State whether the audio was recorded or synthesized and whether the complete line was reviewed. Follow the applicable permissions for the person’s image and voice.
For an Imagild project, use the same separation to prepare assets, while checking the selected workflow’s real input options. This article does not claim that every Imagild video entry accepts Avatar 2.0’s inputs. If a route cannot accept the approved speech, choose a supported production path rather than moving the script into an unrelated prompt field.