🎯 Flow Match Presets

Every other scheduler node in this pack asks you to know something - the right shift value, the right step count, the right CFG for your model. ZImagePresets is the one for when you don't want to know that, you just want to pick your model from a list and get numbers that work. It's a dropdown, not a formula.

What it actually does

Instead of feeding it steps and shift yourself, you pick a preset describing the model and step count you're targeting, and the node returns three things at once: the sigma schedule, the step count that pairs with it, and a recommended CFG. That's the node description in a nutshell - "one-click presets for common models. Returns sigmas, optimal step count, and recommended CFG." Under the hood it's presumably just picking the same shift values the other nodes in this pack expose manually (linear for Z-Image-Turbo, a fixed 3.0 for SD3, a resolution-aware value for Flux), but you don't have to remember any of that - you just tell it which model and step count you're running.

The six presets it ships (per the node's own choices) are Z-Image-Turbo (8 steps), Z-Image-Turbo (9 steps), Flux Dev (20 steps), Flux Schnell (4 steps), SD3 (28 steps), and SD3.5 Turbo (4 steps) - a reasonable cross-section of what's actually being run in the wild right now, distilled and non-distilled both.

The inputs and outputs that matter

• preset (required, enum) - pick your model + step-count combo from the list above. This is the only thing you strictly need to set.
• width / height (optional INT, default 1024, range 64–8192, step 8) - these only do real work for the Flux presets, where the node's description says shift is computed dynamically based on resolution rather than fixed. For Z-Image-Turbo and SD3 presets, where the underlying schedule is linear or fixed-shift, these are along for the ride.

Three outputs come back:

• sigmas (SIGMAS) - plug straight into your SamplerCustom/SamplerCustomAdvanced node, same as every other scheduler in this pack.
• steps (INT) - the step count baked into the chosen preset, useful if you want your workflow to display or log it rather than hardcoding a number that could drift out of sync with the sigmas.
• recommended_cfg (FLOAT) - the CFG the preset expects. For Z-Image-Turbo that's 1.0 per the node's own description (it's guidance-distilled, so anything above 1.0 just doubles render time and burns the image rather than helping).

How to install it

Via ComfyUI Manager: search "ComfyUI-ZImageTurbo-FlowSampler," install, restart. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/capitan01R/ComfyUI-ZImageTurbo-FlowSampler

Restart afterward. No extra Python dependencies, no models to download - same as the rest of the pack.

Common issues & troubleshooting

Treating recommended_cfg as something you have to manually copy over. It's an output value, not a suggestion in a tooltip - if your CFG widget in the sampler chain isn't wired to accept it, you'll need to read the number and set your CFG field to match yourself, or pass it through a conversion node if your ComfyUI version supports widget-to-input conversion on the CFG field.

Picking a preset that doesn't match your actual checkpoint. This node trusts you to know what model you're running. Selecting "Flux Dev (20 steps)" while actually sampling on Z-Image-Turbo will hand you a schedule shaped for the wrong distillation entirely - you'll get worse results than any of this pack's Z-Image-specific nodes, and the failure mode won't be obvious from the error console.

Ignoring width/height on a Flux preset and wondering why results seem generic. Those fields aren't decorative there - Flux's shift genuinely depends on resolution, so leaving them at the 1024 default when you're actually generating a very different aspect ratio or size will hand you a mistuned schedule.