AIHub Expose Frame

AIHub Expose Frame is the pack's answer to frame-by-frame video editing. Where AIHubExposeImage pulls a whole layer or canvas, this node asks the external editor for one specific frame of the video you're working on, then hands it to ComfyUI as a plain IMAGE tensor ready for img2img, an upscale, a ControlNet pass, whatever you do to a single frame before writing it back to the timeline.

It sits inside the same AIHub architecture as everything else in this pack: the value isn't fetched by typing in the web UI. An AIHub client - in this case something like a video editor plugin speaking the protocol over the pack's websocket on port 8111 - connects, tells the node which frame it means, uploads it, and the server injects the file path as a hidden input. Run it bare in the web UI and it'll complain about a missing local_file; that's the protocol working as designed, not a crash.

The inputs that matter

Two inputs do the real work:

• frame_index_type - relative_to_current (default) or absolute. Relative means the client's playhead position plus your frame_index offset; absolute means the literal frame number in the video. If you're building "denoise 3 frames ahead of the playhead," relative is what you want; if you're doing "always fix frame 400," use absolute.
• frame_index - the index itself. Handy detail from the tooltip: if it's out of bounds it clamps to the nearest valid frame rather than erroring. So requesting frame 9999 in a 120-frame clip quietly gives you frame 119. Worth knowing before you debug "why is it always the last frame."

Plus the usual protocol plumbing: id (unique, alphanumeric), label, tooltip, index (sort order), and optional - set true and a missing frame returns zeros instead of failing the run.

Outputs

You get IMAGE, plus the frame's WIDTH and HEIGHT and two genuinely useful ints: FRAME_INDEX (the actual, absolute frame you ended up with) and TOTAL_FRAMES (how long the video is). The README makes a point of this: the outputs are always positive absolute indices even if the client asked with a negative offset - which is exactly what you want when you're about to composite the result back into a timeline and need to know which slot it belongs in.

Install & gotchas

Install through ComfyUI Manager (search ComfyUI-aihub-workflow-exposer) or:

cd ComfyUI/custom_nodes
git clone https://github.com/otavanopisto/ComfyUI-aihub-workflow-exposer

then restart. No requirements.txt, no extra pip packages - it only uses what ComfyUI bundles. The ComfyUI/aihub/... folders should exist and the workflow needs an AIHubWorkflowController with a unique id.

Two traps worth naming. First: because out-of-bounds frames clamp silently, set your frame_index_type deliberately - a stray absolute on a node that meant "three frames from the playhead" will look fine and be subtly wrong. Second: this node is a single frame gateway. If you need a whole batch of frames as a tensor (say, for video-to-video generation or lora training), that's AIHubExposeImageBatch's job, not this node's. Expose Frame is for the frame-by-frame editing flow - one frame in, one frame out.