RunningHub MiniMax H3 FL2VA First / First+Last (Legacy)

This is the node you reach for when you have an image and you want MiniMax H3 to make it move. It packages a first frame - and optionally a last frame - into the MINIMAX_H3_FL_KEYFRAMES object that FL2VA generation runs on. First frame only: H3 animates forward from your image, inventing the trajectory. First + last: you bookend the clip and the model has to connect the dots, which gives you real control over where the motion lands.

Inputs

• first_frame - required, an IMAGE. This is your starting frame, and it sets the composition H3 builds from. Character, product shot, whatever - this is the anchor.
• last_frame - optional, an IMAGE. Connect it and you've gone from "animate from here" to "animate from here to there." This is the higher-control mode, and it's where a lot of the "motion control like Kling" capability the model got praised for comes from.

Output: keyframes (MINIMAX_H3_FL_KEYFRAMES), which feeds the FL2VA Target node (and through it the FL2VA Encode node).

Why this beats prompt-only

H3 is omni-modal, so the keyframe isn't just a style reference - the video VAE encodes it into the actual latent space the sampler works in. The image is doing real structural work, not nudging a prompt. That's why an H3 "first frame → video with synced audio" run can hold a face or a product across the whole clip in a way text alone never could. And because audio is generated jointly, even a dead-simple first-frame run comes back with sound.

The practical pattern

Keep the first and last frames the same aspect ratio as your target, or H3 will have to reconcile mismatched shapes. For character consistency, the same face in both frames goes a long way; for pure motion control, dramatically different poses in first and last is the fun part. Then chain keyframes → RHMiniMaxH3FL2VATarget → RHMiniMaxH3FL2VAEncode → Dual Sigma Sampler → Decode AV.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/RH-RunningHub/ComfyUI-RH-MiniMax-H3.git
pip install -r ComfyUI-RH-MiniMax-H3/requirements.txt

Restart, and pull the ~95 GiB INT8 ConvRot bundle into ComfyUI/models/MiniMax-H3-INT8-CONVROT/ (hf download Gluttony10/MiniMax-H3-INT8-CONVROT --local-dir ./models/MiniMax-H3-INT8-CONVROT, ModelScope in China). Legacy node, still fully functional - modern workflows fold it into RHMiniMaxH3VideoGen. License as pack-wide: H3's Community License excludes the US, EU, UK and South Korea from local-weights use.