bobo-fun

bobo-fun is a project that collects interesting AI open-source projects that run on a Mac (All-In-One).

All code has been tested on a MacBook Pro (M4 Max / 128GB RAM).

News:

• Support QwenLM/Qwen2.5-Omni
• Support HiDream-ai/HiDream-I1
• Support Wan-Video/Wan2.1

Todo:

• Support HunyuanVideo-I2V

Environment

To set up our environment, please run:

conda create --name bobo python=3.11
conda activate bobo

pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cpu

pip install -r requirements.txt

brew install ffmpeg

Usage

Just run you want as belows

commands

• mflux

python run_flux.py

• HiDream

pip install git+https://github.com/huggingface/diffusers.git
python run_hidream.py

• glm-4-9b-chat

python run_glm.py

• Phi-3.5-mini-instruct

python run_phi.py

• DeepSeek-R1-Distill

python run_deepseek_r1.py

• OmniGen

python run_omnigen.py

• StoryDiffusion

python run_story_diffusion.py

• FlowEdit

python run_flow_edit.py

• RMBG

PYTORCH_ENABLE_MPS_FALLBACK=1 python run_rmbg.py

• CogVideo

python run_cogvideo.py

• CogVideoX-Fun

PYTORCH_ENABLE_MPS_FALLBACK=1 python run_cogvideofun.py

• LTX-Video with STGuidance

python run_ltxvideo.py

• mochi ---------> PS. black video bug

python run_mochi.py

• HunyuanVideo

python run_hyvideo.py

• Wan-Video

PYTORCH_ENABLE_MPS_FALLBACK=1 python run_wan_video.py

• Tencent/MimicMotion

conda remove pytorch torchvision torchaudio
conda install pytorch==2.5.1 torchvision==0.20.1 torchaudio==2.5.1 -c pytorch
pip install -r requirements.txt
python run_mimic_motion.py

• xinntao/Real-ESRGAN

python run_realesrgan.py

• hzwer/ECCV2022-RIFE

python run_rife.py

• thewh1teagle/kokoro-onnx

python run_kokoro_82m.py

• suno-ai/bark

python run_suno_bark.py

• hkchengrex/MMAudio

python run_mmaudio.py

• QwenLM/Qwen2.5-Omni

pip install git+https://github.com/huggingface/transformers@v4.51.3-Qwen2.5-Omni-preview
huggingface-cli login (for downloading model)
python run_qwen_omni.py

Issues:

1. 目前几乎所有文生视频项目, 当分辨率或生成帧数较大时, 都有 total bytes of NDArray > 2**32 或 Invalid buffer size 报错问题。这似乎是 mac 本身内部实现问题, 暂无太好方法。

2. Mochi 目前有纯黑视频 bug。

   Mochi: sample_model() ---> model_fn() ---> out_cond 和 out_uncond 模型采样值为 tensor([[nan, nan, ..., nan, nan]...]), 导致纯黑问题。

3. HiDream 生成的图片呈现奇怪的“五彩斑斓”, 有待研究原因

Supports:

• 1. THUDM/glm-4-9b-chat (文生文)

• 2. microsoft/Phi-3.5-mini-instruct (文生文)

• 3. deepseek-ai/DeepSeek-R1-Distill (文生文)

• 4. flux.1 (文生图)

• 5. HiDream (文生图)

• 6. VectorSpaceLab/OmniGen (文生图)

• 7. HVision-NKU/StoryDiffusion (文生图, 擅长故事)

• 8. fallenshock/FlowEdit (修图)

• 9. briaai/RMBG-2.0 (图片背景剔除)

• 10. THUDM/CogVideo (文生视频)

• 11. aigc-apps/CogVideoX-Fun (文生视频)

• 12. Lightricks/LTX-Video (文生视频)

• 13. genmoai/mochi (文生视频)

• 14. Tencent/HunyuanVideo (文生视频)

• 15. Wan-Video/Wan2.1 (文生视频)

• 16. Tencent/MimicMotion (动作/跳舞)

• 17. xinntao/Real-ESRGAN (视频超分)

• 18. hzwer/ECCV2022-RIFE (视频插帧)

• 19. thewh1teagle/kokoro-onnx (文生音频)

• 20. suno-ai/bark (文生音频)

• 21. hkchengrex/MMAudio (文生音频, 擅长音效/音乐)

• 22. QwenLM/Qwen2.5-Omni (多模态)