MiniMax CN I2V Config

MiniMaxCNI2VConfig is the most flexible video config node in MediaHub, and it gets there with one mode input that unlocks three different generation styles. For a single node, that's a lot of surface: reference-image animation, first/last-frame video, and subject-reference video. If you're only going to add one MiniMax video node to a workflow, this is the one.

Output is a MEDIAHUB_VIDEO_REQUEST, and the pipeline is the pack standard: config → MediaHubSubmitVideo → MediaHubResolveVideo → SaveVideo.

The mode input

• i2v - classic image-to-video. Wire first_frame_image, the model animates forward from it. This is the default mental model for "make my picture move."
• fl2v - first-and-last-frame video. Give first_frame_image and last_frame_image and the model invents the middle. This is how you get a clip that starts at A and ends at B, which makes it the closest thing here to controlled "loop-ish" structure.
• s2v - subject-reference video. Feed subject_image and the model keeps that subject consistent while the scene does what the prompt says. Same character-consistency idea as the I2I image node, applied across time.

The optional image inputs (first_frame_image, last_frame_image, subject_image) align to whatever mode you pick - an s2v run ignores a first frame, so match your wires to your mode or the provider will quietly do something you didn't intend.

Inputs that matter

• mode - i2v / fl2v / s2v (above). The whole node hangs off this.
• model - seven choices, the biggest list in the pack: MiniMax-Hailuo-2.3, MiniMax-Hailuo-2.3-Fast, MiniMax-Hailuo-02, I2V-01-Director, I2V-01-live, I2V-01, and S2V-01. The Fast and -live variants are your iteration speeds; the Director is for camera-work prompts; S2V-01 pairs with s2v mode.
• prompt - the motion/action description.
• resolution - 768P / 1080P.
• duration - 6 or 10 seconds.
• prompt_optimizer (default on), fast_pretreatment (off), aigc_watermark (off) - same as T2V.

Output: request → MediaHubSubmitVideo.

Installing it

cd ComfyUI/custom_nodes
git clone https://github.com/vantang/ComfyUI-MediaHub.git
/path/to/ComfyUI/python -m pip install -r ComfyUI-MediaHub/requirements.txt

Restart ComfyUI; under MediaHub/Provider/MiniMax CN/Video. Requirements are requests/Pillow/numpy; you supply a MiniMax CN key (mainland endpoint/account). Source images are JPEG-encoded to base64 before upload, same as every image-input node here.

Common gotchas

Mode/image mismatches are the top mistake - selecting fl2v without a last frame will fail validation or produce garbage, and s2v with only a first frame gives you an i2v that thinks it's something else. The source JPEG round-trip also means the cleaner your input image, the better the motion sticks. And as always, don't stop at Submit - without ResolveVideo polling and downloading, there's no MP4 at the end of the wire.