中文

短剧工作台 (comfyui-AICG3D) plugin package

短剧工作台 (comfyui-AICG3D) is a deeply integrated plugin package.

Verified

Core facts

Read by a script from the official API, not rewritten by a model.

License
GPL-3.0
Languages
Python
Maintainer
JGRFW
Verified

What it is

短剧工作台 (comfyui-AICG3D) is a deeply integrated plugin package. It integrates the core generation capability of MiniMax H3, intelligent segment parsing and advanced rendering strategies in one package. This plugin has a built-in resource library, segmented prompt parsing and H3 core nodes. This plugin does not require installing ComfyUI-MiniMaxH3-Easy or the standalone ComfyUI-AICG3D-PromptSplit separately.

Main features

  1. The plugin supports automatically splitting long scripts/prompts into logical segments.

  2. The plugin has a built-in latent upscaling algorithm designed for H3.

  3. The plugin supports an advanced self-lift sampling scheme.

  4. In the fixed six-role voice library of v1.1.23, girl, boy, female lead, male lead, old woman and old man each have an independent voice seat.

  5. The v1.1.6 update summary states that five H3 task modes are connected to the execution chain.

  6. In v1.1.13, the loader dropdown groups items by folders such as MiniMax, Krea2, Qwen and LTX2.5.

Getting started

  1. For quick installation, copy the comfyui-AICG3D folder into the custom_nodes directory of ComfyUI.

  2. The start-service step is to restart ComfyUI.

  3. Find the AICG3D category in the right-click menu to start creating.

  4. According to the v1.1.11 update summary, after release the plugin can be found by searching AICG3D, 短剧工作台 or MiniMax H3.

  5. A normal installation handles Python dependencies automatically.

  6. After installation, restart ComfyUI and press Ctrl+F5 in the browser.

From the README · powershell
python.exe -m pip install -r requirements.txt
From the README · text
ComfyUI/
└─ models/
   └─ tts/                         # Windows 下 models\TTS 也兼容
      └─ Spark-TTS-0.5B/
         ├─ config.yaml
         ├─ BiCodec/
         ├─ LLM/
         └─ wav2vec2-large-xlsr-53/

Who it is for

The final composite video of the workflow 13 long-video workflow requires FFmpeg. The plugin prefers imageio-ffmpeg, then looks for ffmpeg.exe in the system PATH. The plugin has built-in Spark-TTS inference source, node registration and 5 Step-Audio-speakers reference audios.

For Spark-TTS, only the large model weights need to be downloaded separately. The environment variable AICG3D_SPARK_TTS_MODEL can also point to the model directory. The workflow 13 version 1.2 uses DMADSampler and DMADSigmas.

The v1.1.2 update summary says not to install the standalone ComfyUI-AICG3D-PromptSplit at the same time. In v1.1.23, when the voice library node is not added, the workflow automatically clears invalid voice links and skips voice processing. v1.1.12 batch-checks whether images, audio and video exist after a workflow is opened.

In v1.1.16, audio/video trimming supports the input formats 5, 5 秒, 5s or 0:05. In v1.1.4, when the external sampler and Sigmas are not connected, the original sampler_name, scheduler, steps and denoise continue to be used. In v1.1.5, the Advanced Acceleration, first-pass and second-pass upscale renderers each add optional external sampler and external scheduler (Sigmas) inputs.

v1.1.1 adds the Add next segment and Delete this segment functions. v1.1.15 adds the MiniMax H3 Aicg 3D character white model node (video to motion skeleton) and the segment sampler node. In v1.1.14, non-SafeTensors latent upscale models are loaded with weights_only=True.

In v1.1.24, each second-pass level performs real upscaling and sampling, not video pass-through. The project is open-sourced under GNU GPL v3.0 or later (GPL-3.0-or-later).

Sources

This page is written from the official sources below. Open them to check the original text.

Verified ·

This page is written by AI from the official sources above and checked automatically. This site has not tested it. Found a mistake? Tell us.