H3 Prompt IDE:
H3PromptIDE, displayed in ComfyUI as H3 Prompt IDE, is a rich prompt editor for writing and organizing text for MiniMax H3 workflows while remaining compatible with ordinary text-based workflows. It provides H3-specific authoring assistance through strict mode schemas, contextual completion, reference tokens, and automatic MiniMax H3 Edit instruction context. These features help you compose prompts with references to media and instructions suited to the task you are preparing. Reference media is managed separately through H3 Reference Inputs, keeping the editing experience distinct from media processing. The node itself does not generate images or videos, analyze reference media, or call a model to enhance your writing; its execution simply returns the text you have prepared. This makes it useful when you want a specialized editing environment without introducing an additional generation or prompt-enhancement step.
H3 Prompt IDE Input Parameters:
prompt
A required STRING containing the text you write in the editor. This is a multiline field, so you can organize longer descriptions and instructions across multiple lines rather than placing everything on one line. Its default value is an empty string, and the schema specifies no minimum length, maximum length, or fixed list of choices. The field is marked as supporting ComfyUI dynamic prompts; however, the node’s execution method does not implement its own dynamic-prompt expansion. At execution time, the node returns the received text as a string without adding instructions, rewriting your wording, or analyzing its meaning. Empty text produces an empty string rather than a node-specific validation error. For example, you could write “A slow camera push toward a rain-covered window, with warm interior lighting” and pass that text to a compatible downstream prompt field.
references
An optional H3_PROMPT_REFERENCES connection intended to come from H3 Reference Inputs. It supplies authoring information for the editor’s reference-token palette and previews, helping you refer to connected pictures, video frames, and audio while composing your prompt. This is a connection rather than a text field, so it has no numeric minimum, maximum, or selectable options; no explicit schema default is specified, and the execution method accepts an omitted connection as None. The connection uses a raw link, which allows authoring assistance without evaluating reference media when only text is queued. Most importantly, the execution method discards this argument: it does not inspect the media, append reference information to your prompt, or carry media into the returned text. Reference tokens you actually include in your written prompt remain part of that text, but the associated media must be handled separately by the workflow.
H3 Prompt IDE Output Parameters:
text
A STRING containing the prompt received by the node, returned as ordinary ComfyUI text. This makes the result suitable for downstream inputs that accept a STRING, without requiring a special H3-only text format or connection type. The execution method converts the received prompt to a string and returns an empty string if the prompt is empty; it does not generate additional descriptions, attach media, or perform model-based prompt enhancement. Any reference labels written into the prompt are therefore text, not embedded images, videos, or audio. Connecting this output directly to TextEncodeH3Edit.prompt activates the editor’s task-aware Edit instruction category, providing authoring context for that specific connection while retaining the ordinary STRING output contract.
H3 Prompt IDE Usage Tips:
- Find the node under
text/H3 Prompt IDE, or search for “H3 Prompt IDE,” “rich text prompt,” “prompt editor,” or “minimax h3 prompt editor.” - Use the multiline
promptfield to separate scene descriptions, motion instructions, and editing requests into readable sections. The execution method does not rewrite these sections, so review your wording before sending it downstream. - Connect H3 Reference Inputs to
referenceswhen you need reference-token choices and previews while writing. Leave it disconnected for text-only work; the connection is optional. - Keep reference media connected to the appropriate downstream generation or editing nodes as required by those nodes. Connecting media to the editor’s reference helper does not automatically supply it to a model.
- When using H3 Reference Inputs, plan around its documented reference limits: up to nine pictures, up to three reference video-frame batches, and up to six emitted H3 audio labels. Audio labels account for video soundtracks first, followed by standalone audio. These limits belong to the reference helper, not to the length of your prompt.
- For H3 Edit authoring, connect
textdirectly toTextEncodeH3Edit.promptto activate the task-aware Edit instruction category. A connection through an intermediate node is not documented as providing the same automatic context. - Use the node as an editor rather than a prompt enhancer. If you need MiniMax’s model-based prompt enhancement, the separately provided MiniMax H3 Context IR (Prompt Enhancer) node serves that purpose; its instructions require attaching the same media to the downstream H3 video node in the same order.
- Check that
promptcontains meaningful text before queuing a workflow. This node accepts an empty prompt, so any requirement for nonempty text must be satisfied for the downstream task.
H3 Prompt IDE Common Errors and Solutions:
Empty text reaches the next node
- Explanation: This is a possible workflow symptom, not a custom error message defined by H3PromptIDE. The default prompt is empty, and the execution method returns an empty string when no text is provided. The node does not enforce a nonempty prompt.
- Solution: Enter your intended description in
promptand confirm thattextis connected to the downstream prompt field. If the prompt is supplied through a connection, check the connected source as well.
Reference images, videos, or audio are missing from the generation
- Explanation: The
referencesconnection supports authoring tools and previews only. H3PromptIDE explicitly ignores it during execution and returns text rather than media. - Solution: Connect the actual media to the downstream generation or editing node according to that node’s requirements. Keep the reference-helper connection for composing your prompt, but do not treat it as a replacement for model media connections.
Cannot connect an image, video, or audio directly to references
- Explanation: The
referencessocket expects the custom H3_PROMPT_REFERENCES type, not an individual media connection. The provided source does not specify an exact connection-error message. - Solution: Add H3 Reference Inputs, connect your media to its appropriate sockets, and connect its
referencesoutput to H3PromptIDE’sreferencesinput.
H3 Edit instruction context does not activate
- Explanation: Automatic task-aware Edit context is documented for a direct connection from this node’s
textoutput toTextEncodeH3Edit.prompt. Other destinations or intermediate connections are not documented as activating it. - Solution: Connect
textdirectly toTextEncodeH3Edit.prompt. If you need an intermediate text-processing node, do not assume that the editor will retain the same automatic Edit context.
The prompt is not automatically enhanced or rewritten
- Explanation: This is expected behavior rather than an execution error. H3PromptIDE’s execution method returns your prompt as text; contextual completion and reference tools are authoring features, not a model-based enhancement service.
- Solution: Revise the text in the editor yourself, or use a separate prompt-enhancement node such as MiniMax H3 Context IR if your workflow requires that functionality.
Reference labels describe the wrong media
- Explanation: A reference token in the returned text does not carry the corresponding media with it. If the downstream workflow uses different media or a different reference order, the prompt can refer to the wrong item.
- Solution: Compare the reference labels in your prompt with the media supplied downstream and keep their ordering consistent with the receiving node’s requirements. For Context IR workflows, attach the same media in the same order as documented by that node.
Reference sockets stop growing or additional labels are unavailable
- Explanation: The companion H3 Reference Inputs node has documented limits of nine pictures, three reference video-frame batches, and six emitted H3 audio labels. Video soundtracks count toward the audio-label ordering before standalone audio.
- Solution: Reduce or reorganize your references to fit those limits. When checking available audio labels, include any labels associated with video soundtracks rather than counting only standalone audio connections.
