Best Context Window (UTK)

Some video models won't take just any frame count - they want windows of 4n+1 frames, and they want your clip chopped into matching segments so the temporal conditioning lines up. Figuring out "my 130-frame clip needs what window size to avoid padding waste?" by hand is the kind of arithmetic you'll definitely botch at 1 a.m. Best Context Window (UTK) is that arithmetic in a node: feed it your total frames and it returns the cleanest window size, the padding, and how many segments to split into.

The inputs

• total_frames - how many frames your clip actually has.
• min_window_frames / max_window_frames (defaults 61 / 81) - the range of acceptable window sizes. The node searches only inside this band.

What it computes

It generates all candidate window sizes in [min, max] that satisfy the 4n+1 rule (so 61, 65, 69, …), then for each one works out the arithmetic of splitting your total into windows: how many segments (k = ceil(total / window)), how many padding frames that requires to fill the last segment (padding = k*window − total), and the padded total.

The output is the candidate that minimizes padding - and on a tie, the larger window (larger windows mean fewer, more manageable segments).

Outputs

Four integers, and each one is a wire you can plug somewhere real:

• best_window - the window size to use for your video model's context.
• padding - how many duplicate/padding frames you'll need to reach a clean multiple. Zero is the happy case.
• padded_total - total_frames + padding, i.e. the actual frame count the model will process.
• segments - how many chunks your clip splits into.

So the workflow is: total_frames → this node → feed best_window and segments into your context/latent setup, and know exactly how much padding your pipeline needs instead of discovering it via a failed generation.

Installing it

Standard pack install - ComfyUI Manager → search ComfyUI-UniversalToolkit, or:

cd ComfyUI/custom_nodes
git clone https://github.com/whmc76/ComfyUI-UniversalToolkit
cd ComfyUI-UniversalToolkit && pip install -r requirements.txt

Then restart. It's pure Python integer math - no dependencies, no GPU, no models. It's as close to a free node as this pack has.

Gotchas

• If your min/max range contains no valid 4n+1 value, it falls back to the closest valid window below max rather than erroring - so it never breaks, but double-check the output if you set weird bounds.
• "Best" means least padding, not best video quality. A window that needs zero padding is mathematically clean but might be a size your model handles poorly. The node optimizes the bookkeeping; you still pick the bounds that make sense for the model.
• It only answers "what window," not "how to build the segments." You still wire the frame chopping yourself - this node just makes the numbers trustworthy.

It's a tiny math helper, the kind you'd normally do in a spreadsheet. But if your video pipeline is finicky about context windows, having this in the graph turns "I think 65 works?" into a number you can trust.