StreamDiffusion Fast Sampler

This is the smallest sampler in the whole pack, and that's the point. Once StreamDiffusion Warmup has locked in the negative prompt, steps, CFG, delta, and seed, this node's whole job per call is: take a positive prompt, take a count, produce an image. Two inputs that actually vary, nothing else - the leanest possible interface for the tightest loop you can build in this pack, which matters a lot when you're trying to squeeze latency out of a live generation cycle running many times a second.

If you need your negative prompt or sampling settings to change between frames, this isn't the node for you - that's what StreamDiffusion Sampler (paired with Create Stream directly, no Warmup step) is for. Fast Sampler trades that flexibility for speed, on the assumption that only the subject of your prompt is moving frame to frame while everything else about the generation holds steady.

The inputs and outputs that matter

• warmup_stream (required) - from StreamDiffusion Warmup. This is a distinct type from the plain STREAM that StreamDiffusion Sampler consumes; the two aren't interchangeable.
• positive (required, multiline) - the one thing that's expected to change call to call.
• num (default 1) - how many images per call.

One output: IMAGE.

How to install it

ComfyUI Manager → Install Custom Nodes → search ComfyUI-Diffusers, or manually:

cd ComfyUI/custom_nodes
git clone https://github.com/Limitex/ComfyUI-Diffusers.git
cd ComfyUI-Diffusers
pip install -r requirements.txt

This node is meaningless without a full StreamDiffusion setup behind it, so follow through on the rest of the README:

git clone https://github.com/cumulo-autumn/StreamDiffusion.git
python -m streamdiffusion.tools.install-tensorrt

Given the whole point of this node is squeezing out latency, the TensorRT step is worth actually doing here rather than skipping - it's the difference between "fast" and "as close to real-time as this pack gets." And to keep the graph re-running automatically for continuous generation, turn on Auto Queue in ComfyUI's Extra options.

Common issues & troubleshooting

"Missing node type: StreamDiffusionFastSampler." Standard fix: install the pack via Manager or git clone, restart.

Only fires once instead of looping. Same trap as the regular StreamDiffusion Sampler - Auto Queue needs to be enabled in Extra options for continuous generation. Without it, this node just runs once per manual queue like anything else.

warmup_stream input won't connect. It only accepts the WARMUP_STREAM output from StreamDiffusion Warmup - not the plain STREAM output from StreamDiffusion Create Stream. If you're trying to plug Create Stream directly into this node, that's the wrong pairing; either add a Warmup node in between, or switch to the regular StreamDiffusion Sampler if you actually need per-call control over the negative prompt and sampling settings.

Need to change the negative prompt but it's not exposed here. By design - negative prompt is fixed at whatever StreamDiffusion Warmup was set to when you built the stream. Changing it means going back to the Warmup node, updating it, and re-running; there's no live override on this sampler. If your negative prompt genuinely needs to move frame to frame, this two-node Warmup/Fast Sampler pairing isn't the right tool - use StreamDiffusion Sampler directly instead.