沐阳 H3 · 任意位置关键帧

Sometimes you don't want the model to wander through a whole shot hoping it lands where you want - you want it to pass through a specific pose at a specific moment. That's the whole pitch of 沐阳 H3 · 任意位置关键帧 (H3AnchorKeyframe): you give it an image and a frame index, and H3 treats that frame as a hard constraint to hit. It's the sharp tool in the Myang anchor kit, and it composes with the other one.

How it works is satisfyingly small. The node takes your conditioning, the latent, a vae, your image, and a frame_index. It encodes the image down to a single temporal block and injects it into the conditioning as a keyframe at that index. H3's native multi-keyframe mechanism - which this pack patches into ComfyUI's MiniMaxH3 layout code - then does the work of making the generated video land on that frame. Because it's just conditioning surgery, you can chain as many of these as you like, and you can stack them on top of 沐阳 H3 · 段间多关键帧 (H3AnchorContext) if you want both a continuity seam and a mid-shot pose pin in the same segment. That combination is the pack's "multi-keyframe" story: the seam anchors the head, the keyframes steer the middle.

Inputs and the one rule

• conditioning / latent / vae - the standard trio from your conditioning node.
• image - the pose to pin. It gets resized to the segment's resolution before encoding.
• frame_index - 0-based, and the tooltip spells out the only real rule: it must be less than the total frame count of this segment. The node validates this and raises rather than silently clamping, so you'll know.

Output is a single conditioning socket with the keyframe appended - wire it straight into your sampler. Since it appends rather than replaces, running this node twice in a row gives you two pins, which is the chaining story.

Install and practical notes

Standard pack install: ComfyUI Manager search "ComfyUI-MiniMaxH3-Myang", or git clone https://github.com/civilcoco/ComfyUI-MiniMaxH3-Myang into custom_nodes, then restart. No extra Python dependencies; you supply the H3 model and VAEs.

Honest advice: this is a control node, not a fix-it node. Pinning a frame tightens composition, but it doesn't rescue a shot whose prompt is fighting the pose - and each pin eats reference-encoding cost, so resist the urge to carpet-bomb a segment with ten of them. Start with one pin at the frame you care about (a beat change, an action point) and only add more if H3 keeps missing it. And the standing reminder applies here too: H3's community license geofences the local weights out of the US, EU, UK and South Korea, so check the terms before building a project on it.