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 > H3 Prompt IDE

ComfyUI Extension: H3 Prompt IDE

Repo Name

ComfyUI-H3-Prompt-IDE

Author
ethanfel (Account age: 3465 days)
Nodes
View all nodes(1)
Latest Updated
2026-09-23
Github Stars
0.04K

How to Install H3 Prompt IDE

Install this extension via the ComfyUI Manager by searching for H3 Prompt IDE
  • 1. Click the Manager button in the main menu
  • 2. Select Custom Nodes Manager button
  • 3. Enter H3 Prompt IDE 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

H3 Prompt IDE Description

H3 Prompt IDE is a ComfyUI editor for MiniMax H3 prompts, offering rich and plain-text editing, smart suggestions, media reference previews, structure validation, and task-aware Edit templates.

ComfyUI-H3-Prompt-IDE Introduction

H3 Prompt IDE — rich H3 prompt authoring for ComfyUI

ComfyUI-H3-Prompt-IDE is a VS Code-inspired prompt editor designed for MiniMax H3 workflows in ComfyUI. It gives you a more organized place to write prompts, check their structure, and work with image, video, and audio references.

Instead of managing a long prompt in a basic text box, you can use an editor with:

  • Rich and plain-text views for comfortable writing and precise source inspection.
  • H3-aware suggestions for sections, reference labels, dialogue, and special markers.
  • Reference previews to help you identify the media mentioned in your prompt.
  • Structure checks and diagnostics for supported H3 prompt formats.
  • Task-aware integration with MiniMax H3 Edit, including short starting instructions for editing, character sheets, and scene coverage.
  • A normal ComfyUI STRING output, so your prompt remains ordinary text that can be connected to downstream nodes.

The extension is especially useful when your prompt includes several references, dialogue, or strict H3 formatting. It helps reduce repetitive typing and makes it easier to spot structural problems before generating.

It is an authoring tool, not a generation engine. Your connected H3 workflow still handles models, media conditioning, sampling, and decoding.

How ComfyUI-H3-Prompt-IDE Works

Think of the extension as a script editor with a reference board beside it. You write the creative instructions, the editor helps you organize them, and your H3 generation nodes interpret the finished text.

The two nodes

Find the nodes under text → H3 Prompt IDE.

NodeWhat it does
H3 Prompt IDEProvides the editor, suggestions, section tools, diagnostics, and the final text output.
H3 Reference InputsAssociates connected media with H3 reference labels and previews, keeping media sockets separate from the editor.

The reference node independently expands its available inputs as you connect media:

  • <Picture 1> through <Picture 9>
  • <Video 1> through <Video 3>
  • <Audio 1> through <Audio 6>

These are the editor node’s available reference-label ranges, not a guarantee that every downstream generation workflow accepts that many media items.

A basic authoring workflow

  1. Add H3 Reference Inputs and H3 Prompt IDE.
  2. Connect your image, video-frame, or audio outputs to the matching sockets on H3 Reference Inputs.
  3. Connect its references output to H3 Prompt IDE.
  4. Choose the H3 schema appropriate for your downstream workflow.
  5. Write your prompt using the available sections, suggestions, and reference labels.
  6. Review any diagnostics.
  7. Connect the IDE’s text output to your H3 prompt input.

The text connection looks like this:

H3 Reference Inputs → references → H3 Prompt IDE → text → H3 prompt input

Reference previews do not send media into generation

H3 Reference Inputs handles labels and previews only.

For native Ref2VA generation, connect the same media loaders to the appropriate inputs on your native H3 conditioner as well.

In other words, there are two separate paths:

Media loader → H3 Reference Inputs → editor labels and previews
Media loader → native H3 conditioner → actual generation conditioning

Mentioning <Picture 1> in your prompt is like referring to “reference photo 1” in an art brief: the generation workflow still needs to receive that photo.

Video sockets expect ComfyUI IMAGE frame batches, matching the native H3 ref_videos inputs. They do not expect a generic VIDEO connection.

Native reference naming and audio order

The editor uses H3’s native labels:

<Picture 1>
<Video 1>
<Audio 1>

For example, an instruction might assign identity and clothing separately:

Use <Picture 1> for the character's identity and <Picture 2> for the jacket design.

When assigning audio labels, follow H3’s native presentation order:

  1. Connected video soundtracks first.
  2. Standalone audio afterward.

Keeping the editor’s labels aligned with the conditioner’s media order helps avoid assigning a voice, sound, or visual reference to the wrong item.

ComfyUI-H3-Prompt-IDE Features

Rich and Plain editing

The Plain/Rich source toggle gives you two presentations of your prompt:

  • Rich: a visually organized editing experience with highlighted elements and reference interactions.
  • Plain: direct access to the prompt’s text representation.

Use Rich view when composing and reviewing a reference-heavy prompt. Switch to Plain when you want to inspect exact spelling, brackets, or special tokens.

Both presentations produce a normal text output; Rich view does not require a special downstream prompt format.

H3 schemas, sections, and diagnostics

The editor supports validation for:

  • T2VA
  • I2VA
  • FL2VA
  • L2VA
  • Ref2VA
  • Downstream Edit instructions

A schema is the expected structure of a prompt—similar to a form with particular sections. Choosing the appropriate schema lets the editor check your text against that structure.

Section-insertion tools help you build a prompt without manually retyping every structural element. Diagnostics help you identify formatting issues in the supported H3 structure.

These checks concern prompt structure and supported syntax. They do not guarantee that the model will perfectly follow every creative instruction.

H3-aware completion

Suggestions help you insert H3 elements with less manual typing, including reference labels and supported prompt markers.

The editor supports automatic reference-label spacing, and settings let you control whether completed symbols receive trailing spaces.

For example, a space after a completed <Picture 1> label lets you continue writing naturally:

<Picture 1> defines the room layout.

You can disable automatic suggestions if you prefer a quieter writing environment. Ctrl+Space or Cmd+Space remains available to request completions manually.

Reference and speaker-token editing

Reference and speaker tokens support click-to-replace or delete interactions. This makes it easier to change an assignment without carefully selecting the entire token.

For example, if a wardrobe instruction points to <Picture 2> but should use <Picture 3>, you can replace the reference token rather than rewriting the surrounding sentence.

Marker-replacement interactions are configurable in ComfyUI Settings.

Contextual marker tools

The editor provides contextual insertion and Ctrl/Cmd-click replacement for unstyled:

  • Shot markers
  • Language markers
  • Task-directive markers
  • retention_analysis markers

These tools help you work with structural markers while keeping your attention on the instruction itself.

Dialogue and preferred language

The editor completes, highlights, and validates H3 dialogue tokens, including <d> and </d>.

A dialogue span may look like:

<d>[English] Welcome to the studio.</d>

You can set a preferred dialogue language in ComfyUI Settings. When a language is set, the paired <d>[Language] …</d> span becomes the first matching dialogue completion.

Leave the preference blank if you want the plain dialogue span to appear first.

This setting changes completion priority; it is not an automatic translation feature.

Exact tokenizer-token support

The editor recognizes MiniMax H3’s exact, case-sensitive tokenizer tokens:

<d>
</d>
<|cutoff|>
<|lyrics_start|>
<|lyrics_end|>
<|caption_start|>
<|caption_end|>

The precise spelling matters. For example, existing <cutoff> text remains readable, but the editor recommends the tokenizer-native form:

<|cutoff|>

Correct single-token encoding requires a ComfyUI build containing PR #15808 or its merged equivalent. Editor highlighting alone cannot add that support to an older ComfyUI build.

Customizable editing behavior

Open ComfyUI Settings → H3 Prompt IDE to adjust:

SettingEffect on your editing experience
Default Rich/Plain presentationSets the initial presentation for nodes without a saved per-node choice.
Automatic suggestionsControls whether completions appear while you type.
Preferred dialogue languagePrioritizes the matching language-tagged dialogue completion.
Trailing spaces after completed symbolsControls spacing after inserted completions.
Marker-replacement interactionsControls the relevant replacement behavior.

A saved per-node Rich/Plain choice takes precedence over the default presentation setting.

Text undo and redo

While either editor presentation is focused, native browser text undo and redo are preserved. ComfyUI’s workflow history is prevented from consuming the same shortcut at the same time.

This helps keep “undo my last sentence edit” separate from “undo a change to the node graph.”

Built-in editor-width workaround

The extension includes a scoped workaround for the recurring LiteGraph widget-width bug.

That frontend problem can cause widgets to stop following their node’s width, leaving them clipped, overflowing, or surrounded by unused space. The built-in workaround lets this editor resize correctly without a separate repair node.

ComfyUI-H3-Prompt-IDE and MiniMax H3 Edit

The extension integrates with ComfyUI-MiniMax-H3-Edit, which handles single-image editing, reference-driven generation, directed transformations, character sheets, and scene coverage.

Connect the editor to the Edit encoder

Connect the IDE’s text output directly to the prompt input on TextEncodeH3Edit, displayed as Text Encode H3 Edit / Generate.

The schema selector then shows an Edit category and follows the connected encoder’s task, including:

  • Edit instructions
  • Re-posing
  • Character swaps
  • New camera angles
  • Character-sheet assignments
  • Continuous frozen-scene coverage
  • Cinematic frozen-scene cuts
  • Room + object study, with a compatible Edit version

With MiniMax H3 Edit 0.10.0 or later, connect H3 Edit Options to the encoder and choose its canonical mode. Prompt IDE reads that Options node directly, so changing its mode updates the editor’s context without relying on hidden legacy task/profile values.

Write the instruction, not the full generated wrapper

For supported Edit tasks, you generally write a short, focused instruction. The Edit encoder adds the full H3 task and timing structure.

For example:

Add the black acetate glasses from <Picture 2> to the woman.
Fit them naturally to her face with realistic reflections and contact shadows.
Change nothing else.

A directed camera instruction can be similarly concise:

Arc 45 degrees to camera right at the same height and focal length,
keeping a medium shot.

The editor helps you author the task; the connected Edit encoder handles its detailed execution contract.

Task templates and protection for custom text

When you select a directed task or character-sheet profile, the editor loads a short task-specific starting instruction.

During a live task change:

  • Empty text updates automatically.
  • The previous generated template updates automatically.
  • Custom text triggers a confirmation before replacement.

Loading or switching workflows restores the detected task context without opening a replacement confirmation.

Use Task template whenever you want to restore the current task’s starting instruction.

If the encoder uses use prompt verbatim, the editor returns to your manually selected full H3 schema and does not offer a task template. In that mode, you are responsible for the complete prompt rather than a short instruction.

Frozen-scene camera coverage

For continuous frozen-scene coverage, your instruction should:

  1. Identify the room or scene.
  2. Declare the camera’s orbit center.
  3. Assign optional <Picture N> references as alternate angles or room-design references.

The connected Edit encoder supplies the exact multi-view timing, world-freeze rules, and first/final keyframe alignment.

In semantic generation mode, your instruction can describe a completely new room without manually constructing a full Ref2VA prompt.

Cinematic hard cuts

For cinematic coverage, name one exact visible target: a person, object, architectural feature, or fixed point.

For example:

Coverage target: the woman seated at the desk in <Picture 1>.
Preserve her pose, wardrobe, desk, room geometry, lighting, and shadows
across every cinematic camera cut.

The Edit encoder turns this short instruction into eight timed static shots with true hard cuts and no intermediate camera travel.

Room + object study

This preset requires MiniMax H3 Edit 0.11.0 or later.

Treat the connected pictures as complementary survey views of one physical room, and name one exact object to study.

Important distinctions:

  • Every picture has equal semantic authority.
  • The pictures are interpreted through Qwen visual understanding only, rather than used as input VAE anchors.
  • The legacy source_image socket assigns <Picture 1> its number; it does not make that picture a master image, composition anchor, or timeline keyframe.

The Edit encoder automatically selects a 362-frame semantic reconstruction preset and compiles:

  • Six room-establishing shots.
  • Ten contextual object-study shots.

Each object view is designed to retain the complete object, its floor or rug contact, and recognizable room anchors. The object’s orientation remains fixed while background parallax indicates a change in camera position.

These timing, reconstruction, and camera behaviors belong to the connected Edit encoder, not to Prompt IDE itself.

ComfyUI-H3-Prompt-IDE Models and Compatibility

ComfyUI-H3-Prompt-IDE does not provide or load its own generation models. Its schema choices are prompt-authoring formats, not separate checkpoints.

Your downstream H3 workflow determines which models and conditioning route are required. For example, MiniMax H3 Edit distinguishes FL2VA-based anchored editing and semantic generation from REF2VA-based native reference generation. Selecting a prompt schema in the IDE does not switch those model files for you.

The editor requires a current ComfyUI version with:

  • The V3 node API
  • Autogrow support

For the exact tokenizer tokens listed above, your build also needs the changes from ComfyUI PR #15808 or their merged equivalent.

What is intentionally not included

This standalone editor does not include the Motion Context-specific:

  • Plan, history, or optimizer integration
  • @tag compilation
  • @@@@tag compilation

It uses native H3 reference labels and a normal text output instead. Features described in related extensions—such as AI prompt optimization, segment sampling, or media-library caching—should not be assumed to exist in Prompt IDE.

What’s New with ComfyUI-H3-Prompt-IDE

The available documentation identifies the following version-related integration changes, rather than a complete release-by-release changelog:

Version or requirementDocumented changeWhy it matters
Prompt IDE 0.8.2 and laterReads the connected H3 Edit Options mode and detects the matching Edit task.The instruction template can follow the task you actually selected downstream.
MiniMax H3 Edit 0.10.0 and laterProvides the canonical Options mode integration read directly by Prompt IDE.Hidden legacy task/profile values cannot leave the editor in the wrong task category.
MiniMax H3 Edit 0.11.0 and laterEnables the Room + object study preset.You can author a concise room-survey and target-object instruction while the encoder builds the detailed reconstruction and shot plan.

The current documented feature set also includes task-template protection for custom text, exact tokenizer-token support, configurable completions, reference-token interactions, and the scoped widget-width workaround. No individual release numbers are supplied for those features.

Troubleshooting ComfyUI-H3-Prompt-IDE

My reference appears in the editor, but generation does not use it

  1. Check that the media loader is connected to H3 Reference Inputs.
  2. Check that references is connected to H3 Prompt IDE.
  3. Connect the same media loader to the matching input on the native H3 conditioner.
  4. Confirm that the reference label in your prompt matches the conditioner’s media order.

Why: editor references provide labels and previews, not generation conditioning.

My video output will not connect to a reference socket

  1. Check the output type of your video-loading node.
  2. Use its decoded IMAGE frame-batch output, or a compatible node that supplies one.
  3. Connect that frame batch to the video reference socket.
  4. Supply the appropriate frame-batch connection to the native H3 conditioner as well.

The reference video sockets use the same IMAGE batch representation as native H3 ref_videos, not a generic VIDEO object.

The wrong audio reference seems to be assigned

  1. List the soundtracks associated with connected videos.
  2. Assign those first in H3 audio presentation order.
  3. Place standalone audio after the video soundtracks.
  4. Compare your <Audio N> mentions with the conditioner’s actual order.

Do not assume that the first standalone audio file is always <Audio 1> when video soundtracks are also present.

The Edit task or template does not match my workflow

  1. Confirm that the IDE’s text output is connected directly to the Edit encoder’s prompt input.
  2. Use a compatible Prompt IDE version; Options-mode detection is documented for 0.8.2 and later.
  3. With Edit 0.10.0 or later, connect H3 Edit Options to the encoder.
  4. Change the task using that Options node’s canonical mode.
  5. Use Task template if you want to restore the detected task’s starting instruction.

If the encoder is in prompt-verbatim mode, keeping your manually selected full H3 schema is expected.

Suggestions do not appear while I type

  1. Open ComfyUI Settings → H3 Prompt IDE.
  2. Check whether automatic suggestions are enabled.
  3. Try Ctrl+Space or Cmd+Space to request suggestions manually.

Manual completion remains available when automatic suggestions are disabled.

Changing the default view did not change an existing node

  1. Check whether that node already has a saved Rich/Plain choice.
  2. Change its presentation using the node’s own toggle.
  3. Use the global setting for nodes without a saved choice.

The default does not override an existing per-node preference.

A special token is highlighted, but encoding still seems incorrect

  1. Check the token’s exact spelling and capitalization.
  2. Replace legacy <cutoff> text with <|cutoff|> where appropriate.
  3. Confirm that ComfyUI includes PR #15808 or its merged equivalent.

Recognition by the editor and correct encoding by ComfyUI are separate requirements.

The editor is clipped or does not follow the node’s width

The extension already includes a scoped workaround for the known LiteGraph width issue.

  1. Check that you are using a current version of the extension.
  2. Refresh the ComfyUI interface and try resizing the node again.
  3. Compare the behavior with the documented frontend issue.
  4. If the problem persists, use the official custom-node troubleshooting guide.

A separate width-repair node is not normally needed for Prompt IDE’s editor. The included workaround should not be interpreted as a global repair for every other extension’s widgets.

Frequently asked questions

Can this extension generate images or videos by itself?
No. It produces prompt text for a downstream H3 workflow.

Does Rich view change the output into a special data type?
No. The output remains a normal ComfyUI STRING.

Will choosing a schema change my checkpoint or sampler?
No. The schema controls prompt-authoring context and validation, not model loading or sampling.

Can I use Motion Context @tag or @@@@tag syntax here?
This standalone editor does not compile those tags. Use native H3 reference labels.

Will changing an Edit task erase my custom prompt automatically?
During a live task change, custom text prompts a confirmation before replacement. Empty text and the previous generated template can update automatically.

Why are there no task templates in verbatim mode?
Verbatim mode expects you to provide the complete H3 prompt. The editor therefore uses the manually selected full schema rather than a short task instruction.

Learn More about ComfyUI-H3-Prompt-IDE

Troubleshooting and compatibility

Project background

The author adapted this standalone editor from the Rich Scene Prompt Editor developed for ComfyUI-MiniMaxH3-Contex-Loop, removing its Motion Context-specific integration to provide a normal text-output node.

The rich reference presentation, media miniatures, quick-reference interaction, and compact authoring layout were inspired by ComfyUI-MiniMaxH3-Easy. That is a separate generation suite with its own features and workflows.

The scoped width workaround was adapted with permission from ComfyUI-LegacyWidgetWidthFix.

ComfyUI-H3-Prompt-IDE License

ComfyUI-H3-Prompt-IDE is licensed under the GNU General Public License v3.

H3 Prompt IDE 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.

H3 Prompt IDE detailed guide | ComfyUI