Power Prompter (Umbra Lab)

Power Prompter (Umbra Lab) is the one-node version of everything that happens before a sampler: it loads the model, encodes your prompts, builds the empty latent at the right size, resolves the seed, applies any <lora:...> tags in your prompt, and dumps out model, CLIP, VAE, positive, negative, latent, and a pile of metadata. Wire it into a KSampler and the Save node and you've replaced about eight nodes with one.

That's the "one-node entry point for API workflows" framing the author uses, and it's the right lens: this is the node Umbra Studio's own front end and remote workflows lean on, because it collapses the fragile loader-plus-encode-plus-latent chain into a single call with one set of knobs.

How it works

Everything is driven off its widgets, and it's built from machinery you've already met in this pack:

• Model loading - it delegates to Load Checkpoint (Umbra Lab) under the hood, so model_type runs through the same checkpoint / diffusers / diffusion_model / unet / gguf menu, with the same weight_dtype fp8 options. GGUF still needs ComfyUI-GGUF installed.
• Prompt handling - it strips a legacy [x] checkbox prefix from prompt text, and (like the LoRA syntax node) parses <lora:...> tags, applies them, and removes them from the text. It even supports clip_skip (1–12) if your model family needs it.
• Resolution - aspect_ratio is a dropdown of 33 presets spanning SD1.5 (512-based), SDXL (1024-based), and a 1536 series, plus custom with your own width/height. swap_dimensions flips W/H after the preset resolves, and everything snaps to multiples of 8.
• Seed - control_after_generate offers fixed / increment / decrement / randomize (plus legacy true/false/1/0 values), with a per-node counter so increment/decrement step correctly across runs. style_seed_behavior's same_seed_style_cycle pins the seed across style-expanded prompt jobs.

Outputs worth naming: model, clip, vae, positive, negative, empty_latent (the batch of zero latent you feed the KSampler), prompt_text / negative_prompt_text (cleaned), model_name, seed, steps, cfg, sampler_name, scheduler, denoise, width, height, batch_size. Feed the metadata outputs into Save Image (Umbra Lab) and your PNG records the truth.

The inputs that matter

The required list is long (prompt, negative, model_type, the model name fields, weight_dtype, seed, control_after_generate, aspect_ratio, width/height/batch_size/steps/cfg/sampler/scheduler/denoise), but a beginner sets: prompt_text, negative_prompt, checkpoint_name (or model_type + the matching file field), aspect_ratio, seed + control_after_generate. The optional clip/vae inputs are there to override model-only loaders that return None for CLIP/VAE - wire those in for Flux-style stacks.

Installing it

Part of Umbra-Nodes, the ComfyUI companion pack for Umbra Studio (NocturneLabs' open-source local AI creation suite - Power Prompter is Umbra Studio's modular prompt-card system; this node is its ComfyUI bridge). ComfyUI Manager → search "Umbra Nodes", or:

cd ComfyUI/custom_nodes
git clone https://github.com/Nocturne-Ai-Labs/Umbra-Nodes

Restart ComfyUI. No pip deps; GGUF mode needs ComfyUI-GGUF.

Gotchas

Because it owns the seed counter in-process, it recomputes every run when control_after_generate is set to increment/decrement/randomize - expected, not a leak. The long output list means ComfyUI treats it as carrying lots of data; unused outputs are idle wires, harmless. And the aspect-ratio preset names look like sizes but they're presets: pick custom and set width/height yourself if you want something off the menu, because a preset overrides your typed dimensions.