Pt Linspace

torch.linspace is one of those functions you never think about until you need a smooth ramp of values, and then it's the only tool that fits. Pt Linspace brings it into the graph: you give it a start, an end, and how many points you want between them, and it builds the evenly spaced sequence as a tensor. No Python, no loops - that's the whole point of this pack.

How it works

The implementation is exactly torch.linspace(start, end, steps, dtype=...). The one wrinkle is that all three numbers arrive as text - start, end, and steps are STRING inputs, parsed to float/float/int inside the node. The data_type dropdown picks the output dtype from the full menu: float32, float16, bfloat16, float64, then the integer types (uint8, int8, int16, int32, int64).

Two things to internalize about linspace semantics:

• steps is a count, not a step size. start=0, end=10, steps=5 gives you [0, 2.5, 5, 7.5, 10] - five points. If you're used to range(0, 10, 2), that's a different habit, and mixing them up is the most common source of confusion.
• Endpoints are inclusive. Both start and end appear in the output. That's unlike a lot of sequence functions and it catches people out.

Where it fits

Sweeps are the natural use - evenly spaced CFG values, temperature values, scale factors for a parameter study, all as a tensor you can iterate or slice. It's also handy for building coordinate grids and positional sequences, which is exactly the kind of thing you'd feed into the pack's model-building examples. If you want a ramp or a grid, this is the node.

Gotchas

The defaults are empty strings - all three numeric fields start blank, so the node errors until you type something in. Pick a float dtype unless you truly need integers: the even spacing gets truncated when you choose int8/int32, turning your smooth ramp into a staircase. And keep steps sane; a huge count is a huge tensor for no benefit.

Installing

Pt Linspace lives in the HowToSD/ComfyUI-Pt-Wrapper pack under the "Data Analysis" menu - one pack, ~200 nodes. ComfyUI Manager: search ComfyUI-Pt-Wrapper, install, restart. Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/HowToSD/ComfyUI-Pt-Wrapper

The pack's requirements list is heavy (transformers, peft, accelerate, datasets…), but those exist for the training side. Sequence generation needs only PyTorch, which ComfyUI already ships - skip the pip install for math-only use and restart. No models to download.

Shared pack note: everything speaks the custom TENSOR type, not ComfyUI's IMAGE/LATENT. Convert with Pt From Image (Pt From Image Transpose for (b, c, h, w)), and back with Pt To Image.