mudler/LocalAI · error · ValueError

audio input requires a LongCat-Video-Avatar-1.5 model

Error message

audio input requires a LongCat-Video-Avatar-1.5 model

What it means

ValueError from longcat-video generation: the request includes an audio field but the loaded model classified as MODEL_KIND_BASE (LongCat-Video base), which does text-to-video only. Audio-driven avatar generation requires the LongCat-Video-Avatar-1.5 model; the check fires before _generate_base runs and maps to gRPC INVALID_ARGUMENT.

Source

Thrown at backend/python/longcat-video/backend.py:277

            )
            if ignored_params:
                print(
                    f"longcat-video ignoring unknown request param(s): {', '.join(ignored_params)}",
                    file=sys.stderr,
                )

            os.makedirs(os.path.dirname(request.dst) or ".", mode=0o750, exist_ok=True)
            if hasattr(context, "add_callback"):
                context.add_callback(interrupt_if_cancelled)

            if request.start_image and not os.path.isfile(request.start_image):
                raise ValueError("start_image is not a readable staged file")
            if request.num_frames < 0:
                raise ValueError("num_frames must not be negative")

            if self.model_kind == MODEL_KIND_BASE:
                if request.audio:
                    raise ValueError(
                        "audio input requires a LongCat-Video-Avatar-1.5 model"
                    )
                self._generate_base(request, params)
            else:
                self._generate_avatar(request, params, context)

            return backend_pb2.Result(
                message="Video generated successfully", success=True
            )
        except ValueError as err:
            return self._fail(context, grpc.StatusCode.INVALID_ARGUMENT, str(err))
        except Exception as err:
            print(f"Error generating LongCat video: {err}", file=sys.stderr)
            traceback.print_exc()
            return self._fail(
                context,
                grpc.StatusCode.INTERNAL,
                f"LongCat video generation failed: {err}",

View on GitHub (pinned to 44413a9d06)

Solutions

Omit request.audio when using the base LongCat-Video model (text/image-to-video only)
Or load LongCat-Video-Avatar-1.5 as the model when audio-driven generation is needed

Example fix

# before (base model loaded)
req.audio = "/data/staged/voice.wav"

# after
req.ClearField("audio")  # base model: no audio input

Defensive patterns

Strategy: validation

Validate before calling

def validate_request_for_model(req, model_kind: str):
    """model_kind: 'base' or 'avatar' as reported at load time."""
    if model_kind == "base" and req.audio:
        raise ValueError("audio input is only valid with LongCat-Video-Avatar-1.5")
    if model_kind == "avatar" and not req.audio:
        raise ValueError("avatar model requires audio")
    return req

Try / catch

try:
    stub.GenerateVideo(req)
except grpc.RpcError as e:
    details = e.details() or ""
    if "audio input requires" in details:
        req.ClearField("audio")  # downgrade to text-to-video on base model
        stub.GenerateVideo(req)
    else:
        raise

Prevention

Branch your generation flow on the loaded model kind, not one generic request builder
Surface model capabilities (base vs avatar) in your app's model metadata

When it happens

Trigger: Loading model LongCat-Video (base) and sending a generation request with request.audio set; pointing the model config at a base checkpoint while reusing an avatar workflow that sends audio.

Common situations: Switching model repos without updating the client; testing the cheaper base model with an avatar (talking-head) pipeline.

Related errors

AI-assisted analysis of mudler/LocalAI@44413a9d06 (2026-08-15). Data as JSON: /api/errors/168cdd950c066e36. Report an issue: GitHub.