Save 4 hours! We auto-setup your workflow! Free!

Drop your workflow.json — we handle every dependency, custom node, and model. Just open the link and run.

Auto-Setup Workflow Json (Free) Now!
ComfyUI > Nodes > ComfyUI-FeiHou-Easy-H3

ComfyUI Extension: ComfyUI-FeiHou-Easy-H3

Repo Name

ComfyUI-FeiHou-Easy-H3

Author
FX-FeiHou (Account age: 2190 days)
Nodes
View all nodes(1)
Latest Updated
2026-11-12
Github Stars
0.08K

How to Install ComfyUI-FeiHou-Easy-H3

Install this extension via the ComfyUI Manager by searching for ComfyUI-FeiHou-Easy-H3
  • 1. Click the Manager button in the main menu
  • 2. Select Custom Nodes Manager button
  • 3. Enter ComfyUI-FeiHou-Easy-H3 in the search bar
After installation, click the Restart button to restart ComfyUI. Then, manually refresh your browser to clear the cache and access the updated list of nodes.

Visit ComfyUI Online for ready-to-use ComfyUI environment

  • Free trial available
  • 16GB VRAM to 80GB VRAM GPU machines
  • 400+ preloaded models/nodes
  • Freedom to upload custom models/nodes
  • 200+ ready-to-run workflows
  • 100% private workspace with up to 200GB storage
  • Dedicated Support

Run ComfyUI Online

ComfyUI-FeiHou-Easy-H3 Description

ComfyUI-FeiHou-Easy-H3 creates MiniMax H3 videos using an in-node panel for image, video, and audio references, with prompt-linked media and optional LoRA, prompt optimization, and memory controls.

ComfyUI-FeiHou-Easy-H3 Introduction

ComfyUI-FeiHou-Easy-H3 is a set of ComfyUI nodes for creating MiniMax H3 videos. Its main generation node includes a media panel, so you can choose reference pictures, videos, and audio directly inside the node instead of connecting a separate loader for every file.

The extension is designed to make reference-heavy workflows easier to manage. You can keep up to nine pictures, three videos, and three independent audio files in the panel, see picture and video previews, and refer to your chosen media in a prompt. Uploaded files are recorded with the workflow so the panel can restore them when you reopen it.

Beyond media handling, the extension provides controls for LoRA stacks, prompt optimization, memory use, external model-loading workflows, and reusable H3 reference packages called RefMods. You can use the basic media panel without using those more advanced options.

How ComfyUI-FeiHou-Easy-H3 Works

Think of the main node as a workspace for gathering the ingredients of a video. Your prompt describes what should happen; pictures and videos give H3 visual references; audio can provide sound references. The node prepares those inputs for the rest of the ComfyUI generation workflow.

In reference-to-video mode, select media in the node’s slots and use @ in your prompt to pick a reference label such as <Picture 1>, <Video 1>, or <Audio 1>. The label tells the workflow which selected reference you mean. For example, a prompt can describe a scene and point to a particular picture as its visual reference. A video’s existing soundtrack stays paired with that video; the three audio slots are for separate reference audio.

In image or first-and-last-frame mode, the node uses only the first two picture slots. With one picture, an advanced setting determines whether it represents the opening or closing frame. With two pictures, they represent the first and last frames. Switching modes does not erase media selected in the other slots, so you can return to reference mode later.

Prompt optimization is an optional preparation step. When you enable its advanced controls and choose a configured service or local language model, the extension can expand or reinterpret a prompt before generation. The prompt preview node lets you inspect the resulting text rather than guessing what was sent onward.

ComfyUI-FeiHou-Easy-H3 Features

Media Panel and Reference Labels

  • Nine picture slots appear in a fixed 3 × 3 grid, alongside three video slots and three independent audio slots.
  • You can click a slot to choose a file or drag a file into it. Pictures and videos show previews in the node; audio slots show filenames.
  • Selected uploads are recorded with the workflow and can be restored when it is reopened.
  • In reference-to-video mode, type @ to choose numbered picture, video, and audio references for your prompt.

Reference-to-video mode needs at least one picture or video; independent audio alone is not enough. The media-panel limits do not include external RefMods.

Generation and Workflow Nodes

The nodes appear in ComfyUI under FeiHou Easy H3:

NodeWhat it is for
ComfyUI-FeiHou-Easy-H3Main generation node with the embedded media panel.
FeiHou Easy H3 LoaderLoads the H3 model and its text encoder, video VAE, and audio VAE; accepts a LoRA stack and applies it to the model.
加载LoRA(旁路,仅模型)(用于调试)Lets you add, enable, disable, and reorder multiple LoRAs in a stack connected to the Loader.
FeiHou Easy H3 Model AdapterConnects an external standard ComfyUI model-loading chain.
FeiHou Easy H3 OutputExposes separate outputs, including conditioning, latent, video VAE, audio VAE, FPS, and the final prompt. Its reference_map output can help you check reference numbering.
FeiHou Easy H3 提示词预览Displays the final expanded or reverse-generated prompt carried by the H3 context.
FeiHou Easy H3 接续合并Helps join continuing segments by retaining the continuation tail and trimming overlapping video and audio. See the production-pack guide for its connections and controls.

The default output settings are 24 FPS and 10 seconds. Available reference-picture short-edge sizes are 480, 544, 640, 736, 768, 832, 928, 1024, and 1088. These settings give you control over the clip and reference preparation; larger picture settings may demand more memory.

Prompt Optimization

The 🐵Easy H3 group in ComfyUI Settings holds API settings and prompt-optimization rules. Supported service choices include Zhipu, xFlow-API aggregation, Ollama, Alibaba Cloud, DeepSeek, and custom OpenAI-, Gemini-, or Ollama-compatible APIs. You can also create, edit, and select your own prompt plans.

On the main node, prompt optimization requires Advanced Options to be open, an API interface to be configured and selected, and a plan to be selected. A plan has no effect when its API is not enabled. Preset services do not come with an API key or model already supplied. Saved API keys remain in the plugin’s local ComfyUI user configuration and are not exported with workflows.

For artists using a local language model, the extension supports GGUF files, complete Transformers model directories, and standalone .safetensors files with built-in architecture recognition. See the local-model dependency guide for supported configurations and setup details.

Memory Controls

Advanced Options include Force Unload, which reclaims unused cache after sampling and before VAE decoding, and Full Low-VRAM Chunking (Experimental), which processes several H3 components in smaller pieces. The latter may help when available GPU memory is limited, but it is experimental. When it is enabled, do not also chain ModelAttentionBackend / comfy kitchen attention.

RefMod References

The main node and the separate FeiHou Easy H3 Setup accept an optional refmods input from ComfyUI-MiniMaxH3Mod. A RefMod is a reusable reference package. It does not occupy one of the media panel’s picture, video, or audio slots, though it does add reference tokens and memory cost.

In reference-to-video mode, connect the mods output of Load H3 RefMods directly, or use the corresponding output from its Create or Master nodes. Do not also Apply or Text Encode the same reference separately, because that would inject it twice. RefMods created upstream can have their own strength, copies, and modality choices; Easy H3 uses the processed reference it receives. A saved Apply or Step Curve configuration is not automatically applied through this connection.

Enabled RefMods receive labels before ordinary uploaded media, with numbering tracked separately for pictures, videos, and audio. For example, if a video RefMod is <Video 1>, the first video uploaded to the panel becomes <Video 2>. The @ menu can show enabled packages by type, number, and name; reference_map or the logs can show the resulting numbering. A reference with zero strength or one filtered out upstream does not take a number.

RefMod references remain tied to their package source in a saved workflow. If you remove or disable a referenced package, the extension warns that its prompt reference is no longer valid. A visual RefMod can satisfy reference mode without a panel upload, but an audio-only RefMod still needs a picture, video, or visual RefMod. Reference packages can increase memory use, and using several character references does not guarantee that their identities will remain distinct.

Storyboards and Continuing Clips

The experimental production-pack storyboard loader can read a Shotlist from a folder or ZIP package and apply shot-specific prompts, song excerpts, and reference pictures. It uses ComfyUI’s native queue to generate shots in sequence. For longer work made in segments, the continuation-and-merge node works with FeiHou Toolbox Video Combine V2 to save segments and merge the completed video.

ComfyUI-FeiHou-Easy-H3 Models

The extension’s Loader works with FL2VA and REF2VA H3 model paths and loads the accompanying text encoder, video VAE, and audio VAE. Choose the path appropriate to the generation workflow you are building; the original documentation does not provide a model-by-model quality comparison.

For local prompt optimization, the standard edition includes configurations for Qwen3 0.6B, 1.7B, 4B, 8B, 14B, and 32B, plus Mistral 7B Instruct v0.3. These are supported configurations, not a claim that the model files are bundled. Renaming a supported file does not prevent the extension’s built-in architecture recognition. The provided documentation does not specify output-quality or speed results for each size, so choose a model your system can run and compare its prompt results with the preview node.

A LoRA is an optional model adjustment managed through the separate LoRA stack node. You can try combinations by enabling, disabling, and reordering entries before the stack reaches the Loader. A RefMod serves a different purpose: it carries reusable reference media information rather than being a LoRA. RefMod’s Full Reference mode keeps more source detail at a higher reference-token cost; its Compressed Reference mode uses fewer tokens but may lose detail. Neither mode guarantees that only a character’s identity, style, or another desired trait will transfer.

Troubleshooting ComfyUI-FeiHou-Easy-H3

My reference-to-video workflow says it needs a visual reference.

  1. Check that you selected reference-to-video mode.
  2. Add at least one picture or video to the media panel, or connect an effective visual RefMod.
  3. Do not rely on an independent audio slot or an audio-only RefMod as the sole reference.

My prompt plan is selected, but it does not change the prompt.

  1. Open Advanced Options on the main node.
  2. In the 🐵Easy H3 settings, check that an API or supported local-model interface is configured.
  3. Select the enabled interface in the node, then select the prompt plan.
  4. Connect the prompt preview node to inspect the final prompt.

A <Picture i> or <Video i> label points to the wrong source.

  1. Check whether RefMods are connected; they are numbered before panel uploads within each media type.
  2. Use the @ menu to select the intended named reference again.
  3. Inspect FeiHou Easy H3 Output.reference_map or the workflow logs to confirm the numbering. If an older prompt label was created with a temporary upload, remove that old reference and select the RefMod from the current menu.

I removed a RefMod and now its prompt reference is invalid.

  1. Reconnect or re-enable the intended package, if you still need it.
  2. Otherwise, remove the invalid prompt reference.
  3. Choose an available reference from the @ menu. The warning prevents an old package label from silently attaching to a different upload.

I run out of GPU memory.

  1. Try a smaller reference-picture short-edge size or fewer references.
  2. Consider Force Unload in Advanced Options.
  3. If needed, test Full Low-VRAM Chunking (Experimental) without also chaining ModelAttentionBackend / comfy kitchen attention.
  4. Remember that external RefMods also add reference-token and memory costs.

A prompt API works locally but not through a remotely hosted ComfyUI page.

The extension limits settings and prompt-optimization routes to access from the machine running ComfyUI by default. Remote access requires the server-side opt-in described in the original project settings. This is an administrator decision: people who can open that ComfyUI page could use already-saved API keys to make optimization requests and consume credits, even though they cannot see the keys themselves.

Do I lose selected media when I change modes?
No. Image or first-and-last-frame mode uses only the first two picture slots, but switching modes does not delete your other selections. They remain available when you return to reference mode.

Does a RefMod guarantee a consistent face or voice?
No. Reference packages can help present reusable material to H3, but identity retention and generation quality need testing. The referenced RefMod documentation specifically does not establish voice cloning as a working result.

Learn More about ComfyUI-FeiHou-Easy-H3

ComfyUI-FeiHou-Easy-H3 Related Nodes

RunComfy
Copyright 2025 RunComfy. All Rights Reserved.

RunComfy is the premier ComfyUI platform, offering ComfyUI online environment and services, along with ComfyUI workflows featuring stunning visuals. RunComfy also provides AI Models, enabling artists to harness the latest AI tools to create incredible art.

ComfyUI-FeiHou-Easy-H3 detailed guide | ComfyUI