ComfyUI-MiniMax-H3-API Introduction
ComfyUI-MiniMax-H3-API adds nodes for creating MiniMax H3 Video V2 videos inside ComfyUI. You can start with a written scene description, guide a video with opening and ending images, or use supported images, videos, and audio as references. The extension also helps you manage generation tasks, download finished videos, and preview them in your workflow.
For example, you could describe a camera move through a painted landscape, supply an image to guide the opening shot, or use several visual references to establish a character and setting. Different MiniMax models offer different combinations of reference support, resolution, and generation duration, so you can choose the node that fits the idea you are developing.
Generation happens through the MiniMax API, which requires a MiniMax account and API key. The key is read when a node runs; it is not placed in saved workflow files or request outputs. API usage may incur charges.
How ComfyUI-MiniMax-H3-API Works
Think of a workflow as a production brief. The Content Builder gathers your prompt and any connected media into the material MiniMax needs. A Generate Video node sends that material to MiniMax as a task. The extension then checks the task until it finishes, and the Preview Video node can download and display the result.
The Content Builder chooses the type of brief from what you connect:
- Prompt only: Creates a text-to-video request. Use this when your written description is enough to guide the scene.
- First frame, last frame, or both: Uses images to guide the start or end of a video. When frame inputs are connected, the builder uses the
adaptiveaspect ratio. - Reference media: Uses connected images, videos, or audio to inform the generation. At least one reference image or video is required; audio cannot be the only reference.
You can connect native ComfyUI IMAGE, VIDEO, and AUDIO inputs. The extension converts them into the format needed for the API. The complete JSON request must be no larger than 64 MB.
Frame-guided and reference-media inputs are different modes: do not mix them in one request. For the Max and Max Turbo nodes, an ending frame also requires a starting frame.
ComfyUI-MiniMax-H3-API Features
Build a video request
MiniMax H3 Content Builder assembles and validates the content sent to MiniMax. Its inputs include first_frame, last_frame, image_1–image_9, video_1–video_3, and audio_1–audio_3. You do not need to select a content mode separately; the connected inputs determine it.
For reference-based Max generation, you can use up to 9 images, 3 videos, and 3 audio clips, with no more than 12 reference media items in total. Each reference video or audio clip must be 2–15 seconds long. Combined reference-video duration cannot exceed 15 seconds, and combined reference-audio duration has its own 15-second limit. MiniMax performs media duration, download, and decoding validation through its API.
Generate and guide videos
- MiniMax H3 Generate Video starts a video task and automatically checks its progress until completion.
- MiniMax H3 Max Generate Video supports text-to-video, first-frame or first-and-last-frame guidance, and image, video, and audio references. Choose
480Por768Pand a duration from 5–15 seconds. - MiniMax H3 Max Turbo Generate Video supports text-to-video and frame-guided generation at
480Por768Pfor 5–15 seconds. It does not support reference images, videos, or audio. - MiniMax H3 Fast Generate Video supports text-to-video, first/last-frame guidance, and multimodal references at
480Pfor 4–15 seconds. The node applies its requiredstyle1subtype automatically. The stated generation cost is ¥0.30 per second.
Both Max nodes offer optional settings:
| Setting | Choices | What it changes |
|---|---|---|
prompt_expansion_mode | disabled, balanced, quality | Controls optional prompt expansion. balanced is the default; disabled leaves expansion off. |
seed | 0 or 1–4294967295 | 0 omits the seed field so the server can choose a random seed. A nonzero value sends your chosen seed. |
Existing Max workflows do not need to add these optional inputs. Polling is always on, and AIGC watermarking is always disabled; neither is a node setting.
Refine, regenerate, and manage tasks
MiniMax H3 Context IR starts an asynchronous prompt-enhancement task and returns enhanced content you can reuse. MiniMax H3 Regenerate 2K takes an exact 768P generation request and its source video to create a higher-resolution version. These features are not supported by Max, Max Turbo, or H3 Fast.
To regenerate a video, connect request_json from Generate Video to generation_request_json on Regenerate 2K, and provide the exact 768P source video. MiniMax requires that video to contain audio and match its H3 768P output specification.
For tasks you have already submitted:
- MiniMax H3 Query Task checks a task created within the last seven days.
- MiniMax H3 List Tasks displays and filters recent tasks.
- MiniMax H3 Cancel Delete Task cancels queued tasks or deletes records for succeeded or failed tasks.
- MiniMax H3 Preview Video downloads a finished result as
video/MiniMax_日期_时间.mp4. Its preview scales with the node’s size.
Understand the Max cost estimate
The Max nodes display a standard-price estimate, not a guaranteed bill. They show CNY for the China endpoint and USD when base_url is https://api.minimax.io. Your account’s actual billing takes precedence. The nodes do not convert charges into credits.
The following rates come from the supplied September 6, 2026 Max / Max Turbo API document:
| Item | China CNY | Overseas USD |
|---|---|---|
Max 480P output, per second | 0.33 | 0.05 |
Max 768P output, per second | 0.50 | 0.08 |
Turbo 480P output, per second | 0.17 | 0.025 |
Turbo 768P output, per second | 0.26 | 0.04 |
Max reference video input, per second, with 480P output | 0.37 | 0.0553 |
Max reference video input, per second, with 768P output | 0.97 | 0.143 |
| Max reference image beyond the first two, per image | 0.50 | 0.074 |
For Max, the estimate adds output duration × output rate, reference-video duration × video rate, and the image rate for each reference image after the first two. Reference audio and prompt expansion have no separate charge in this estimate. If the usage details are incomplete, the node reports that it cannot calculate the full cost rather than guessing from output duration alone.
ComfyUI-MiniMax-H3-API Models
| Model | Use it when you need | Supported options and limits |
|---|---|---|
| MiniMax H3 | The standard H3 generation workflow, including workflows that use H3-Context-IR or 768P-to-2K regeneration. | The supplied documentation does not specify its full resolution and duration choices. |
| MiniMax H3 Max | A choice between written prompts, frame guidance, and image/video/audio references. | 480P or 768P; 5–15 seconds. |
| MiniMax H3 Max Turbo | Text-to-video or frame-guided generation without reference-media inputs. | 480P or 768P; 5–15 seconds. |
| MiniMax H3 Fast | 480P generation with text, frame guidance, or multimodal references. | 480P only; 4–15 seconds; stated cost ¥0.30/second. |
“Fast” is the model’s name; the supplied information does not provide a measured speed comparison. Max and Max Turbo do not use style, H3-Context-IR, or 2K regeneration. H3 Fast also does not support H3-Context-IR or 2K regeneration.
What's New with ComfyUI-MiniMax-H3-API
The supplied information does not include a version-by-version changelog. It does identify these additions and behaviors, without assigning them release numbers:
- Separate Max and Max Turbo nodes: You can choose a node according to whether your workflow needs multimodal references. Max accepts them; Max Turbo does not.
- Optional Max prompt expansion and seed inputs: You can adjust these settings while leaving older Max workflows without the new inputs.
- Regional Max pricing estimates: The nodes show standard-price estimates in CNY or USD according to the configured endpoint, making the potential cost easier to assess before generation.
- Regional endpoint guidance: The supplied September 6, 2026 document specifies
https://api.minimax.cnfor China andhttps://api.minimax.iofor overseas accounts. An explicitly configuredbase_urlremains in effect until you change it.
Troubleshooting ComfyUI-MiniMax-H3-API
A request fails when you use an ending frame.
- Check which generation node you are using.
- For Max or Max Turbo, connect a
first_frameas well as thelast_frame. - Remove any reference-media inputs from that frame-guided request.
A reference-media request is rejected.
- Confirm that at least one reference image or video is connected; audio alone is not valid.
- Check the applicable limits: up to 9 images, 3 videos, 3 audio clips, and 12 reference items total for Max.
- Check that each reference video and audio clip is 2–15 seconds long, and that the combined durations stay within their separate 15-second limits.
- If the API reports a media problem, check that the source can be downloaded and decoded. The API performs those checks.
Reference inputs do not work with Max Turbo.
- Use MiniMax H3 Max Generate Video for image, video, or audio references.
- Use Max Turbo only for a written prompt or supported frame-guided inputs.
A 2K regeneration request does not work.
- Start with a video generated at
768Pusing the applicable H3 generation workflow. - Connect that task’s
request_jsontogeneration_request_jsonon Regenerate 2K. - Provide the exact source video from that request, including its audio.
- Do not use a Max, Max Turbo, or H3 Fast request for this feature.
The task query cannot find an older task.
- Check the task identifier.
- Confirm that the task was created within the last seven days, the period supported by Query Task.
The displayed price looks incomplete or differs from your bill.
- Check whether the node has the output and reference-media usage details needed for an estimate.
- Check whether
base_urlpoints to the endpoint for your account’s region. - Treat the display as a standard-price estimate; your actual account billing prevails.
Frequently asked questions
Will my API key be saved in a shared workflow?
No. The key is read from local configuration when a node executes and is not included in workflow files or request outputs.
Can I combine opening and ending frames with reference images or clips?
No. Frame-guided inputs and reference-media inputs cannot be mixed in the same request.
Can I use audio without an image or video reference?
No. Reference content needs at least one image or video.
Can I control whether the node waits for generation to finish?
Polling is always enabled.
Does the extension add an AIGC watermark?
AIGC watermarking is always disabled in these nodes.
Are example workflows available?
Yes. Workflow JSON files are in examples/, with input notes in examples/README.md. The examples do not contain an API key.
Learn More about ComfyUI-MiniMax-H3-API
The example workflows in examples/ are a useful starting point for seeing how the nodes connect. Read examples/README.md alongside them for required inputs and usage notes.
For details about the MiniMax API operations behind the nodes, see the official documentation:
- Create a video generation task (https://platform.minimaxi.com/docs/api-reference/video-generation-v2-create)
- Query a task (https://platform.minimaxi.com/docs/api-reference/video-generation-v2-query)
- List tasks (https://platform.minimaxi.com/docs/api-reference/video-generation-v2-list)
- Cancel or delete a task (https://platform.minimaxi.com/docs/api-reference/video-generation-v2-delete)
- H3-Context-IR prompt enhancement (https://platform.minimaxi.com/docs/api-reference/video-generation-v2-h3-context-ir)
- Video regeneration (https://platform.minimaxi.com/docs/api-reference/video-generation-v2-regeneration)
