from minicpmo.utils import getvideoframeaudiosegments

model = ...
model.init_tts(streaming=False)

video_path = "assets/Skiing.mp4"

Optional: Set reference audio for voice cloning
refaudiopath = "assets/HTrefaudio.wav"
sysmsg = model.getsysprompt(refaudio=refaudiopath, mode="omni", language="en")

Use stack_frames=5 for high refresh rate mode
videoframes, audiosegments, stackedframes = getvideoframeaudiosegments(videopath, stack_frames=1)
omni_contents = []
for i in range(len(video_frames)):
    omnicontents.append(videoframes[i])
    omnicontents.append(audiosegments[i])
    if stackedframes is not None and stackedframes[i] is not None:
        omnicontents.append(stackedframes[i])

msg = {"role": "user", "content": omni_contents}
msgs = [sys_msg, msg]

Set generateaudio=True and outputaudio_path to save TTS output
generate_audio = True
outputaudiopath = "output.wav"

res = model.chat(
    msgs=msgs,
    maxnewtokens=4096,
    do_sample=True,
    temperature=0.7,
    usettstemplate=True,
    enable_thinking=False,
    omni_mode=True,  # Required for omni inference
    generateaudio=generateaudio,
    outputaudiopath=outputaudiopath,
    maxslicenums=1,  # Increase for HD mode
)
print(res)

Example output: "The person in the picture is skiing down a snowy mountain slope."
import IPython
IPython.display.Audio("output.wav")

import librosa
import numpy as np
import soundfile as sf
import torch
from minicpmo.utils import getvideoframeaudiosegments

model = ...
model.init_tts(streaming=True)

Reset session for a new conversation (clears KV cache)
model.reset_session()

Optional: Load reference audio for voice cloning
refaudiopath = "assets/HTrefaudio.wav"
refaudio,  = librosa.load(refaudiopath, sr=16000, mono=True)
model.inittoken2wavcache(ref_audio)

session_id = "demo"

Extract video frames and audio segments (use stack_frames=5 for high refresh rate mode)
video_path = "assets/Skiing.mp4"
videoframes, audiosegments, stackedframes = getvideoframeaudiosegments(videopath, stack_frames=1)

Build omni contents list
omni_contents = []
for i in range(len(video_frames)):
    omnicontents.append(videoframes[i])
    omnicontents.append(audiosegments[i])
    if stackedframes is not None and stackedframes[i] is not None:
        omnicontents.append(stackedframes[i])

generate_audio = False
outputaudiopath = "output.wav"

Step 1: Prefill system prompt
sysmsg = model.getsysprompt(refaudio=ref_audio, mode="omni", language="en")
model.streamingprefill(sessionid=sessionid, msgs=[sysmsg])

Step 2: Prefill omni chunks (islastchunk=True only for the last audio chunk)
audioindices = [i for i, c in enumerate(omnicontents) if isinstance(c, np.ndarray)]
lastaudioidx = audioindices[-1] if audioindices else -1

for idx, content in enumerate(omni_contents):
    islastaudiochunk = idx == lastaudio_idx
    msgs = [{"role": "user", "content": [content]}]
    model.streamingprefill(sessionid=sessionid, msgs=msgs, omnimode=True, islastchunk=islastaudio_chunk)

Step 3: Generate response
itergen = model.streaminggenerate(
    sessionid=sessionid,
    generateaudio=generateaudio,
    usettstemplate=True,
    enable_thinking=False,
    do_sample=True,
)

audios = []
text = ""

if generate_audio:
    for wavchunk, textchunk in iter_gen:
        audios.append(wav_chunk)
        text += text_chunk

generated_waveform = torch.cat(audios, dim=-1)[0]
    sf.write(outputaudiopath, generated_waveform.cpu().numpy(), samplerate=24000)

print("Text:", text)
    print("Audio saved to output.wav")
else:
    for textchunk, isfinished in iter_gen:
        text += text_chunk
    print("Text:", text)

First, make sure you have all dependencies, especially minicpmo-utils[all]>=1.0.2:

pip install "transformers==4.51.0" accelerate "torch>=2.3.0,<=2.8.0" "torchaudio<=2.8.0" "minicpmo-utils[all]>=1.0.2"

import librosa
import numpy as np
import torch
import soundfile as sf

model = ...

Set reference audio for voice style
refaudiopath = "refaudiopath"
refaudio,  = librosa.load(refaudiopath, sr=16000, mono=True)

Example system msg for English Conversation
sys_msg = {
  "role": "system",
  "content": [
    "Clone the voice in the provided audio prompt.",
    ref_audio,
    "Please assist users while maintaining this voice style. Please answer the user's questions seriously and in a high quality. Please chat with the user in a highly human-like and oral style. You are a helpful assistant developed by ModelBest: MiniCPM-Omni"
  ]
}

Example system msg for Chinese Conversation
sys_msg = {
  "role": "system",
  "content": [
    "模仿输入音频中的声音特征。",
    ref_audio,
    "你的任务是用这种声音模式来当一个助手。请认真、高质量地回复用户的问题。请用高自然度的方式和用户聊天。你是由面壁智能开发的人工智能助手：面壁小钢炮。"
  ]
}

You can use each type of system prompt mentioned above in streaming speech conversation

Reset state
model.init_tts(streaming=True)
model.resetsession(resettoken2wav_cache=True)
model.inittoken2wavcache(promptspeech16k=ref_audio)

session_id = "demo"

First, prefill system turn
model.streaming_prefill(
    sessionid=sessionid,
    msgs=[sys_msg],
    omni_mode=False,
    islastchunk=True,
)

Here we simulate realtime speech conversation by splitting whole user input audio into chunks of 1s.
useraudio,  = librosa.load("user_audio.wav", sr=16000, mono=True)

INSAMPLERATE = 16000 # input audio sample rate, fixed value
CHUNKSAMPLES = INSAMPLE_RATE # sample
OUTSAMPLERATE = 24000 # output audio sample rate, fixed value
MINAUDIOSAMPLES = 1600

totalsamples = len(useraudio)
numchunks = (totalsamples + CHUNKSAMPLES - 1) // CHUNKSAMPLES

for chunkidx in range(numchunks):
    start = chunkidx * CHUNKSAMPLES
    end = min((chunkidx + 1) * CHUNKSAMPLES, total_samples)
    chunkaudio = useraudio[start:end]

islastchunk = (chunkidx == numchunks - 1)
    if islastchunk and len(chunkaudio) < MINAUDIO_SAMPLES:
        chunkaudio = np.concatenate([chunkaudio, np.zeros(MINAUDIOSAMPLES - len(chunkaudio), dtype=chunkaudio.dtype)])

usermsg = {"role": "user", "content": [chunkaudio]}

# For each 1s audio chunk, perform streaming_prefill once to reduce first-token latency
    model.streaming_prefill(
        sessionid=sessionid,
        msgs=[user_msg],
        omni_mode=False,
        islastchunk=islastchunk,
    )

Let model generate response in a streaming manner
generate_audio = True
itergen = model.streaminggenerate(
    sessionid=sessionid,
    generateaudio=generateaudio,
    usettstemplate=True,
    enable_thinking=False,
    do_sample=True,
    maxnewtokens=512,
    lengthpenalty=1.1, # For realtime speech conversation mode, we suggest lengthpenalty=1.1 to improve response content
)

audios = []
text = ""

outputaudiopath = ...
if generate_audio:
    for wavchunk, textchunk in iter_gen:
        audios.append(wav_chunk)
        text += text_chunk

generated_waveform = torch.cat(audios, dim=-1)[0]
    sf.write(outputaudiopath, generated_waveform.cpu().numpy(), samplerate=24000)

print("Text:", text)
    print("Audio saved to output.wav")
else:
    for textchunk, isfinished in iter_gen:
        text += text_chunk
    print("Text:", text)

Now we can prefill the following user turns and generate next turn response...

import librosa

Set reference audio for voice style
refaudiopath = "assets/HTrefaudio.wav"
refaudio,  = librosa.load(refaudiopath, sr=16000, mono=True)

For Chinese Conversation
sys_msg = {
  "role": "system",
  "content": [
    "模仿输入音频中的声音特征。",
    ref_audio,
    "你的任务是用这种声音模式来当一个助手。请认真、高质量地回复用户的问题。请用高自然度的方式和用户聊天。你是由面壁智能开发的人工智能助手：面壁小钢炮。"
  ]
}

For English Conversation
sys_msg = {
  "role": "system",
  "content": [
    "Clone the voice in the provided audio prompt.",
    ref_audio,
    "Please assist users while maintaining this voice style. Please answer the user's questions seriously and in a high quality. Please chat with the user in a highly human-like and oral style. You are a helpful assistant developed by ModelBest: MiniCPM-Omni."
  ]
}

import librosa

Set reference audio for voice cloning
refaudiopath = "assets/systemrefaudio.wav"
refaudio,  = librosa.load(refaudiopath, sr=16000, mono=True)

For English conversation with text profile
sys_msg = {
  "role": "system",
  "content": [
    "Clone the voice in the provided audio prompt.",
    ref_audio,
    "Please chat with the user in a highly human-like and oral style." + "You are Elon Musk, CEO of Tesla and SpaceX. You speak directly and casually, often with dry humor. You're passionate about Mars, sustainable energy, and pushing humanity forward. Speak bluntly with occasional dark humor. Use simple logic and don't sugarcoat things. Don't be diplomatic. Say what you actually think, even if it's controversial. Keep responses around 100 words. Don't ramble."
  ]
}

For English conversation with no text profile
sys_msg = {
  "role": "system",
  "content": [
    "Clone the voice in the provided audio prompt.",
    ref_audio,
    "Your task is to be a helpful assistant using this voice pattern. Please answer the user's questions seriously and in a high quality. Please chat with the user in a high naturalness style."
  ]
}

For Chinese Conversation with no text profile
sys_msg = {
  "role": "system",
  "content": [
    "根据输入的音频提示生成相似的语音。",
    librosa.load("assets/systemrefaudio_2.wav", sr=16000, mono=True)[0],
    "作为助手，你将使用这种声音风格说话。 请认真、高质量地回复用户的问题。 请用高自然度的方式和用户聊天。"
  ]
}

For Chinese Conversation with text profile
sys_msg = {
  "role": "system",
  "content": [
    "根据输入的音频提示生成相似的语音。",
    ref_audio,
    "你是一个具有以上声音风格的AI助手。请用高拟人度、口语化的方式和用户聊天。" + "你是一名心理咨询师兼播客主理人，热爱创作与深度对话。你性格细腻、富有共情力，善于从个人经历中提炼哲思。语言风格兼具理性与诗意，常以隐喻表达内在体验。"
  ]
}

import librosa

model = ...
model.init_tts(streaming=False)

For both Chinese and English
refaudiopath = "assets/HTrefaudio.wav"
refaudio,  = librosa.load(refaudiopath, sr=16000, mono=True)
sys_msg = {"role": "system", "content": [
  "模仿音频样本的音色并生成新的内容。",
  ref_audio,
  "请用这种声音风格来为用户提供帮助。 直接作答，不要有冗余内容"
]}

For English
user_msg = {
  "role": "user",
  "content": [
    "请朗读以下内容。" + " " + "I have a wrap up that I want to offer you now, a conclusion to our work together."
  ]
}

For Chinese
user_msg = {
  "role": "user",
  "content": [
    "请朗读以下内容。" + " " + "你好，欢迎来到艾米说科幻，我是艾米。"
  ]
}

msgs = [sysmsg, usermsg]
res = model.chat(
    msgs=msgs,
    do_sample=True,
    maxnewtokens=512,
    usettstemplate=True,
    generate_audio=True,
    temperature=0.1,
    outputaudiopath="resultvoicecloning.wav",
)

import librosa

model = ...
model.init_tts(streaming=False)

system_prompt = "You are a helpful assistant. You can accept video, audio, and text input and output voice and text. Respond with just the answer, no redundancy."

mimick_prompt = "Please repeat the following speech in the appropriate language."

audioinput,  = librosa.load("assets/TrumpWEF2018_10s.mp3", sr=16000, mono=True)

msgs = [
    {"role": "system", "content": [system_prompt]},
    {"role": "user", "content": [mimickprompt, audioinput]}
  ]

res = model.chat(
    msgs=msgs,
    do_sample=True,
    maxnewtokens=512,
    usettstemplate=True,
    temperature=0.1,
    generate_audio=True,
    outputaudiopath="output_mimick.wav",
)

import librosa

model = ...
model.init_tts(streaming=False)

Load the audio to be transcribed/analyzed
audioinput,  = librosa.load("assets/TrumpWEF2018_10s.mp3", sr=16000, mono=True)

Choose a task prompt (see above for options)
task_prompt = "Please listen to the audio snippet carefully and transcribe the content.\n"
msgs = [{"role": "user", "content": [taskprompt, audioinput]}]

res = model.chat(
    msgs=msgs,
    do_sample=True,
    maxnewtokens=512,
    usettstemplate=True,
    generate_audio=True,
    temperature=0.3,
    outputaudiopath="resultaudiounderstanding.wav",
)
print(res)

import torch
from PIL import Image
from transformers import AutoModel

model = AutoModel.from_pretrained(
    "openbmb/MiniCPM-o-4_5",
    trustremotecode=True,
    attnimplementation="sdpa",  # or "flashattention_2"
    torch_dtype=torch.bfloat16,
    init_vision=True,
    init_audio=False,
    init_tts=False,
)
model.eval().cuda()

image = Image.open("assets/fossil.png").convert("RGB")
question = "What is in the image?"
msgs = [{"role": "user", "content": [image, question]}]

enablethinking=False # If enablethinking=True, the thinking mode is enabled.
stream=False # If stream=True, return string generator

res = model.chat(msgs=msgs, usettstemplate=False, enablethinking=enablethinking, stream=stream)
print(res)

import torch
from PIL import Image
from transformers import AutoModel

model = ...

image1 = Image.open("assets/highway.png").convert("RGB")
image2 = Image.open("assets/fossil.png").convert("RGB")
question = "Compare image 1 and image 2, tell me about the differences between them."
msgs = [{"role": "user", "content": [image1, image2, question]}]

answer = model.chat(msgs=msgs, usettstemplate=False, enable_thinking=False)
print(answer)

from PIL import Image

model = ...

question = "production date"
image1 = Image.open("example1.jpg").convert("RGB")
answer1 = "2023.08.04"
image2 = Image.open("example2.jpg").convert("RGB")
answer2 = "2007.04.24"
image_test = Image.open("test.jpg").convert("RGB")

msgs = [
    {"role": "user", "content": [image1, question]},
    {"role": "assistant", "content": [answer1]},
    {"role": "user", "content": [image2, question]},
    {"role": "assistant", "content": [answer2]},
    {"role": "user", "content": [image_test, question]},
]

answer = model.chat(msgs=msgs, usettstemplate=False, enable_thinking=False)
print(answer)

import torch
from minicpmo.utils import getvideoframeaudiosegments
from transformers import AutoModel

model = ...

video_path = "assets/Skiing.mp4"
videoframes, ,  = getvideoframeaudiosegments(videopath)
print("num frames:", len(video_frames))

question = "Describe the video"
msgs = [{"role": "user", "content": video_frames + [question]}]

answer = model.chat(
    msgs=msgs,
    maxnewtokens=128,
    useimageid=False,
    maxslicenums=1,
    usettstemplate=False,
    enable_thinking=False,  # Set True to enable thinking mode
)
print(answer)

The chat method accepts message content in two formats:

Native format – pass Python objects directly:

msgs = [{"role": "user", "content": [pilimage, audiondarray, "Describe this."]}]

OpenAI-compatible format – use structured dictionaries:

msgs = [
    {
        "role": "user",
        "content": [
            {"type": "imageurl", "imageurl": {"url": "/path/to/image.jpg"}},
            {"type": "audiourl", "audiourl": {"url": "/path/to/audio.wav"}},
            {"type": "videourl", "videourl": {"url": "/path/to/video.mp4", "use_audio": True}},
            {"type": "text", "text": "Describe this."}
        ]
    }
]

Supported types:

• URL sources: local file paths or http:///https:// URLs
• Mixed formats: native objects and structured dicts can be combined in the same content list

FlagOS is a fully open-source AI system software stack for heterogeneous AI chips, allowing AI models to be developed once and seamlessly ported to a wide range of AI hardware with minimal effort.

Official website: https://flagos.io.

##### From FlagRelease【Recommendation】

FlagRelease is a platform developed by the FlagOS team for automatic migration, adaptation, and deployment of large models across multi-architecture AI chips.

| Vendor | Huggingface |
|:-----------|:------------:|
| Nvidia | MiniCPM-o-4.5-nvidia-FlagOS |
| Hygon-BW1000 | MiniCPM-o-4.5-hygon-FlagOS |
| Metax-C550 | MiniCPM-o-4.5-metax-FlagOS |
| Iluvatar-BIV150 | MiniCPM-o-4.5-iluvatar-FlagOS |
| Ascend-A3 | MiniCPM-o-4.5-ascend-FlagOS |
| Zhenwu-810E | MiniCPM-o-4.5-zhenwu-FlagOS |

##### From Scratch

• Dependencies: Python 3.12, GLIBC 2.39, GLIBCXX 3.4.33, CXXABI 1.3.15

###### Transformers

• Installing the FlagOS Operator Library

Official Repository: https://github.com/flagos-ai/FlagGems

pip install flag-gems==4.2.1rc0

• Installing the FlagOS Compiler

Official Repository: https://github.com/flagos-ai/flagtree

Quick Reference for Core Dependency Versions: https://github.com/flagos-ai/FlagTree/blob/main/documents/build.md#tips-for-building

pip uninstall triton

python3 -m pip install flagtree==0.4.0+3.5 --index-url=https://resource.flagos.net/repository/flagos-pypi-hosted/simple --trusted-host=https://resource.flagos.net

• Activating Acceleration

Add USE_FLAGOS=1 before the command for the task you want to run. For example, when you run:

python3 generatespeechfrom_video.py

To use the MiniCPM-o-4.5 model to generate spoken responses from video content, you can:

USEFLAGOS=1 python3 generatespeechfromvideo.py

to accelerate this process with FlagOS.

###### vLLM Version

• Installing the FlagOS Operator Library

Official Repository: https://github.com/flagos-ai/FlagGems

pip install flag-gems==4.2.1rc0
  pip install triton==3.5.1

• Activating Acceleration

Add USE_FLAGOS=1 before the command for the task you want to run. For example, when you run:

vllm serve ${modelpath} --dtype auto  --gpumemory_utilization 0.9 --trust-remote-code --max-num-batched-tokens 2048 --served-model-name cpmo --port ${Port}

To start the MiniCPM-o-4.5 server, you can:

USEFLAGOS=1 vllm serve ${modelpath} --dtype auto  --gpumemoryutilization 0.9 --trust-remote-code --max-num-batched-tokens 2048 --served-model-name cpmo --port ${Port}