Kling 2.6 Pro Motion MCP guide for image-to-video

Preflight the source image before reserving credits

For local stdio, keep the file inside the configured upload root, which defaults to the process working directory. Published `@offlinecreator/mcp` 0.1.2 accepts `.png`, `.jpg`, `.jpeg`, `.webp`, or `.gif`, rejects empty and non-regular files, caps the decoded image at 10 MiB, and checks the file signature instead of trusting the extension. Base64 input is checked against the same byte limit and recognized image signatures.

This is an MCP upload contract, not a statement that every visually valid image is suitable creative input. Confirm that you have permission to use the source, remove sensitive material that should not reach a cloud provider, and review the actual frame rather than trusting its filename. Remote MCP cannot read a local path through the published client; use the supported base64 path when the remote transport needs image bytes.

Path rejected

Move the file under the upload rootDo not broaden the root merely to expose an unrelated directory.

Type rejected

Use PNG, JPEG, WebP, or GIFRenaming an unsupported file does not pass the magic-byte check.

Size rejected

Keep decoded input at or below 10 MiBResize or recompress an authorized working copy before retrying.

Duration and ratio controls: do not infer unsupported knobs

The underlying fal Kling v2.6 Pro image-to-video documentation requires a prompt and start-image URL. It documents 5 seconds as the default duration and allows 5- or 10-second provider requests. That provider capability does not mean both choices are exposed by OfflineCreator MCP.

In published `@offlinecreator/mcp` 0.1.2, neither `generate` nor `upload_input` has a duration argument. The package exposes 16:9, 9:16, and 1:1 in shared tool schemas, but a broad enum is not proof that Kling 2.6 Pro Motion accepts every value. Current repository model data is more specific than those public surfaces, but it is not a reader-accessible citation for the deployed service. Require the connected `list_models` response or a public model-specific Studio record before presenting duration or ratios as current Studio facts.

Provider duration

5 seconds by default; provider schema also lists 10 secondsThis describes fal's direct endpoint, not an MCP duration selector.

MCP duration input

Not exposed in package 0.1.2Do not add an invented duration property to either tool call.

Shared MCP ratio enum

16:9, 9:16, or 1:1Request syntax only; Kling Motion compatibility is not established by this draft.

What Kling 2.6 Pro Motion does in Studio

Reader-accessible OfflineCreator pages list Kling 2.6 Pro Motion as an image-to-video entry for image animation, disclose fal as the routed provider, and currently price the entry at 42 credits. Refresh the connected catalog before reserving credits because model availability and prices can change.

Published `@offlinecreator/mcp` 0.1.2 tells `generate` callers to use an ID from `list_models`, but the reviewed reader-accessible pages and package files do not bind a Kling Motion Studio ID to the named entry. Stop unless the connected catalog returns the exact image-to-video entry, ID, and current charge. The public package establishes the discovery and upload workflow, not a complete model-specific constraint record.

Workflow

Image-to-videoA source image is required after the generation is reserved.

Current Studio cost

42 credits per generationCurrent catalog price; use list_models before spending.

Provider boundary

Routed through falStudio is a cloud workflow, not on-device processing.

Reserve, upload, submit, then wait

Published `@offlinecreator/mcp` 0.1.2 documents a two-tool handoff for an image-to-video model that the connected catalog actually returns. First call `generate` with that returned model ID and a motion prompt; include a ratio only if current Studio evidence binds it to the exact entry. The implementation reserves an image-to-video generation and returns an upload requirement instead of immediately sending the job to the provider.

Next call `upload_input` with that generation ID and either a local `filePath` or `imageBase64`. The tool uploads the image and submits the reserved job. Set `wait` there if the client should poll after submission, or use `wait_generation` separately. Do not place image bytes in the creative prompt and do not assume that `wait: true` on the first call bypasses the required upload.

1. Discover

Call list_modelsUse the returned model ID and current credit cost.

2. Reserve

Call generateProvide the motion direction and retain the returned generation ID.

3. Attach and submit

Call upload_inputSend the approved image for the reserved generation, then poll or wait.

Review motion against the approved still

Write the prompt around movement that is not already visible: identify one camera move, the subject action, and restrained environmental motion. For example, an approved bottle still might ask for a slow push-in, a slight turntable rotation, and a soft background-light change. This is an editorial prompt pattern, not a measured recipe or a guarantee of adherence.

After completion, compare the first frame and several later frames with the approved source. Check silhouette, logos and readable text, product geometry, faces, hands, background edges, lighting direction, and any unintended crop. Reject drift that changes the approved meaning. This research pass did not run a fresh generation or matched-output test, so it makes no quality, speed, identity-preservation, or reliability claim.

Choose the next guide by what is still unknown

Use the model directory to refresh the live catalog, the duration guide to compare only documented clip lengths, or the source-image guide to prepare an input before reserving a run. These links answer different next questions rather than repeating this page's model query.

Scope and evidence boundary

This page owns the model-specific implementation question “kling motion mcp”: the public package's reserve, local-input validation, upload, submit, and review sequence for the catalog-listed 42-credit entry. The broader `/models/image-to-video-model` page owns workflow selection between image-to-video and text-to-video. This page does not document Kling's separate Motion Control product, direct fal billing, or text-to-video prompting. Current public surfaces still do not expose a model-specific Studio ID, fixed Studio duration, or confirmed ratio set. No runtime was observed. Keep the page researched, unreviewed, draft, and consolidated into the parent until those gaps are independently closed.