Codex + Claude · MCP tutorial · MiniMax H3

Generate MiniMax H3 image-to-video with Codex or Claude MCP

Turn one source image into a motion-rich video without writing an API client. Codex or Claude handles the upload, selects MiniMax H3 in the Medux tool, submits the render, monitors progress, and returns the finished clip.

Input: JPG, PNG, or WebPModel: MiniMax H3Output: generated video
Codex and Claude connected through Medux MCP for MiniMax H3 image-to-video generation
One Medux MCP workflow, available from both Codex and Claude, with MiniMax H3 selected for image-to-video generation.

Workflow at a glance

Your agent orchestrates the work. Medux manages the upload and task lifecycle. MiniMax H3 generates the video.

1. Connect Medux MCP
2. Choose a source image
3. Upload and keep file ID
4. Select MiniMax H3
5. Submit and monitor
6. Verify the output

Before you start

Medux accessA Medux account and API key with enough plan access and credits for the selected MiniMax H3 render settings.
Agent clientCodex or Claude Code with Streamable HTTP MCP support and permission to call Medux tools.
Source imageA clear JPG, PNG, or WebP. Use a stable composition, visible subject, and enough room for the motion you intend to create.
Motion directionA concise prompt that names subject action, camera movement, pacing, atmosphere, and details that should remain unchanged.
New-model check. Before submitting, have the agent inspect the live medux_video_generate_image_to_video definition and use the stable model_id whose display name is MiniMax H3. Do not accept the default model merely because the tool preselects one.

Connect Medux MCP

Use the same Medux remote server and API key in either client. Keep the real key out of screenshots, repositories, and prompts you plan to share.

Codex

Add Medux as a Streamable HTTP server

In the Codex app, open Settings → MCP servers → Add server, choose Streamable HTTP, and enter the Medux URL. A reusable configuration can also be stored in Codex config with the key read from an environment variable.

Codex config.toml[mcp_servers.medux]
url = "https://mcp.medux.ai"
bearer_token_env_var = "MEDUX_API_KEY"

Restart Codex after saving, then open the MCP server list and confirm Medux is enabled.

Claude Code

Add Medux from the terminal

Use the Medux remote MCP command, replacing the placeholder with your own API key.

Claude Code commandclaude mcp add --transport http medux \
  https://mcp.medux.ai \
  --header "Authorization: Bearer sk-medux-..."

Start a fresh Claude session and confirm that Medux tools are available before continuing.

Copy a complete Codex or Claude MCP prompt

Both versions deliberately require a model checkpoint and an argument review before the generation call.

Codex prompt
In this workspace, use Medux MCP to create an image-to-video clip with MiniMax H3.

1. Find and inspect the source image at {IMAGE_PATH}.
2. If needed, create a Medux upload URL, upload the image, and keep the returned file_id.
3. Inspect the live medux_video_generate_image_to_video tool definition. Select the exact model_id whose display name is MiniMax H3; do not use the default Wan model.
4. Use this motion prompt: {MOTION_PROMPT}
5. Set duration_seconds to {DURATION}, clarity to {CLARITY}, aspect_ratio to {ASPECT_RATIO}, and title to {TITLE}.
6. Show me the source-image mapping and every tool argument before execution.
7. After approval, submit the task, monitor it until completion, download the video to this workspace, and verify playback, requested motion, subject consistency, and composition.
Claude prompt
Using Medux MCP, help me generate a MiniMax H3 video from {IMAGE_PATH}.

Restate the source image, intended motion, duration, clarity, aspect ratio, and output title first. Upload the image and keep its Medux file_id. Inspect the live medux_video_generate_image_to_video tool definition and choose the exact model_id labeled MiniMax H3 rather than the default model.

Motion prompt: {MOTION_PROMPT}
Duration: {DURATION} seconds
Clarity: {CLARITY}
Aspect ratio: {ASPECT_RATIO}
Title: {TITLE}

Summarize the proposed tool arguments and ask for approval. When the task finishes, return the generated video and a checklist confirming playback, requested motion, subject consistency, and composition.

Step-by-step guide

01

Prepare one strong first frame

The image becomes the visual starting point for the generated clip. Prefer a clean subject silhouette, readable depth, and a composition that leaves space in the direction of movement.

  • Use JPG, PNG, or WebP and keep the source image available locally.
  • Avoid asking for motion that immediately moves the subject outside the frame.
  • If identity matters, describe what must stay stable in the motion prompt.
02

Write the motion, not a second still image

Describe how the scene changes over time. Name the subject action, camera behavior, pace, environmental motion, and preservation constraints.

Example motion promptThe woman turns slightly toward the camera and smiles while her hair moves naturally in a soft breeze. Use a slow cinematic dolly-in, preserve her facial identity, clothing, lighting, and background composition, and avoid sudden cuts or extra objects.
03

Let the agent upload the image

Ask Codex or Claude to create a Medux upload URL and upload the local file. Keep the local filename-to-file_id mapping visible so you can confirm that the correct first frame is passed to generation.

Approval check: the uploaded asset must be the intended source image, not the tutorial cover, a preview, or another image in the same folder.
04

Resolve the live MiniMax H3 model ID

Model catalogs evolve. Have the agent inspect the tool schema and explicitly report the matching MiniMax H3 identifier before it prepares the request.

Expected argument shape{
  "file_id": "{file_id}",
  "prompt": "{motion_prompt}",
  "model_id": "{LIVE_MINIMAX_H3_MODEL_ID}",
  "duration_seconds": 5,
  "clarity": "720P",
  "aspect_ratio": "16:9",
  "title": "MiniMax H3 image-to-video test"
}

If MiniMax H3 is not present in the live tool definition, stop instead of silently substituting another model; check account access and the current Medux model catalog.

05

Review the Medux request

Compare the proposed call with the current Create Image to Video API reference. Confirm all fields before approval.

POST/api/app/v1/generate/video/image_to_video

MCP tool: medux_video_generate_image_to_video

FieldWhat to verify
file_idMatches the uploaded source image.
promptDescribes motion, camera behavior, pacing, and preservation constraints.
model_idThe live stable identifier for MiniMax H3—not the default model.
duration_secondsA supported duration that matches the intended shot.
clarityA supported output clarity such as 720P or 1080P.
aspect_ratioMatches the target placement, for example 16:9 or 9:16.
titleA searchable task title that distinguishes the render.
06

Submit and monitor the asynchronous task

After approval, Medux returns a task. Keep its ID and let the agent query progress until the task reaches a terminal state.

Task lifecycleQUEUED → PROCESSING → COMPLETED

If the task is still running:
wait briefly, then query the same task_id again.

If the task fails:
report the Medux error, keep the arguments visible,
and do not create repeated paid retries automatically.
07

Download and verify the generated video

Open the returned result URL or ask the agent to download the file into the workspace. Watch the whole clip rather than checking only the first frame.

  • The video opens and plays from start to finish.
  • The requested subject and camera motion are present.
  • Identity, clothing, lighting, and composition remain acceptably consistent.
  • No unexpected objects, hard cuts, or severe edge distortions appear.
  • The output duration, clarity, and aspect ratio match the approved request.

Why use Medux MCP for MiniMax H3?

You keep the entire workflow inside Codex or Claude: local-file discovery, secure upload preparation, model selection, task submission, status checks, and result retrieval. The same prompt pattern remains reusable when you change the source image, motion, aspect ratio, or output target.

One agent workflow
Explicit H3 model check
Auditable arguments
Automatic task monitoring

MiniMax H3 MCP FAQ

Can I use the same Medux tool in both Codex and Claude?

Yes. Both clients call medux_video_generate_image_to_video through the Medux remote MCP server. Their setup and conversation style differ, but the Medux request and task lifecycle are the same.

Why does the tutorial use a live MiniMax H3 model lookup?

New model catalogs and default selections can change. Resolving the entry labeled MiniMax H3 from the live tool definition prevents accidental use of the default image-to-video model and keeps the workflow accurate as Medux updates its catalog.

What should I do if MiniMax H3 is not listed?

Do not substitute another model automatically. Confirm that Medux MCP is current, check your account plan and model access, then consult the linked API reference or Medux support before submitting a paid render.

Which image-to-video settings should I choose?

Start with a short clip and a supported clarity and aspect ratio. Use 16:9 for landscape video or 9:16 for vertical social output, and confirm the current supported values in the API reference before approval.

Compare the same endpoint across agents or continue with adjacent video-generation workflows. For provider routing and fallback beyond MiniMax, read the multi-model image-to-video platform guide.

Create Image to Video API referenceMedux MCP connection overviewImage to Video with Codex MCP and MeduxImage to Video with Claude MCP and MeduxText to Video with Codex MCP and MeduxText to Video with Claude MCP and MeduxOfficial Codex image-to-video workflow on GitHubOfficial Claude image-to-video workflow on GitHub

Quick answer

How do I generate a MiniMax H3 video from an image with Codex or Claude MCP?

Connect Codex or Claude to the Medux remote MCP server, upload a source image, ask the agent to call medux_video_generate_image_to_video with the live MiniMax H3 model ID, approve the arguments, and monitor the asynchronous task until the video is ready.

What does this Medux tutorial cover?

Connect Codex or Claude to the Medux remote MCP server, upload a source image, ask the agent to call medux_video_generate_image_to_video with the live MiniMax H3 model ID, approve the arguments, and monitor the asynchronous task until the video is ready.

Related topics: Create Image to Video with Codex, Claude MCP · Medux Create Image to Video API · Codex, Claude Medux tutorial

Model-specific image-to-video workflow

Run MiniMax H3 image-to-video with Codex or Claude MCP

This model guide targets developers comparing a MiniMax H3 video generator, MiniMax image-to-video API, or agent-driven photo-to-video workflow. It shows how to prepare the source image, route the generation through Medux MCP, review model settings, monitor the task, and validate the returned clip.

Use the general image-to-video tutorial for provider-neutral workflow design, then use this page when MiniMax H3 is the selected model. Keeping model-specific and task-level intent on separate pages avoids competing pages with identical copy.

When should I use a model-specific image-to-video tutorial?

Use it when the query or implementation explicitly names MiniMax H3. Use the provider-neutral image-to-video page when model routing and interchangeable providers matter more.

Production, trust, and developer intent

Run an async, agent-ready workflow to minimax h3 image-to-video

Use Medux as an AI media API for developers when you need to minimax h3 image-to-video from Codex or Claude. The production pattern is submit, keep the task ID, poll status, fetch the result, and verify the output; it supports asynchronous video generation and multi-model media orchestration without hiding model-specific limits.

Before commercial use, verify source rights, the selected model terms, output licensing, brand approvals, and any required disclosure. This is a commercial-use review workflow, not a blanket commercial license.

Should an AI generation API use task polling or a webhook?

This Medux tutorial demonstrates task polling. Use a webhook only when the current API reference explicitly documents callback delivery, authentication, and retry behavior.

Build with the API

API reference for this workflow

Use the endpoint documentation below to automate this Medux workflow directly in your application.