ComfyUI Node
H3 Canvas
The canvas H3 samples at — size and duration — carried on the h3_inputs wire so it can be bypassed or swapped between passes. Snaps the duration UP onto H3's legal frame grid (n % 17 == 5), so 5 seconds is 124 frames, and both axes onto the 32px grid, and reports how far past H3's trained envelope you land. Pick the resolution itself with a Resolution node and wire width and height in.
H3 Canvas
- h3_inputs
- h3_inputs
- report
◄width1344►
◄height768►
◄seconds5.0►
CategoryNynxz/H3
Inputs (4)
| Name | Type | Default | Description |
|---|---|---|---|
| width | INT | 134432–16384 | Canvas width, snapped to the 32px grid — the VAE's /16 then the DiT's 2x2 patch. H3's own envelope is a 768 short edge capped at 768*1344; the report says how far past that you are. |
| height | INT | 76832–16384 | Canvas height, on the same 32px grid. |
| seconds | FLOAT | 5.00.2–150 | Duration at 24 fps, snapped UP to the legal frame grid — the report names what you actually get. H3's trained range is ~5-15s; past that the rope runs off the end of what it saw and sung or spoken content degrades into gibberish well before a minute. |
| h3_inputsopt | NYNXZH3_INPUT | Chain from H3 Image / Video / Audio, or leave unwired. Assets pass through untouched; only the canvas is set. |
Outputs (2)
| Name | Type | Description |
|---|---|---|
| h3_inputs | NYNXZH3_INPUT | Into H3 Encode, or on through more asset nodes. |
| report | STRING | What you actually get: the snapped canvas and frame count, the aspect ratio, and how far past H3's base envelope it lands. For the numbers as values, put an H3 Extras on the wire. |