NovelAIRequestPayload

If NovelAI_Request is the phone call, NovelAI_Request_Payload (shown as NovelAIRequestPayload) is the phone you dial with. It's where every generation setting for NovelAI's hosted models lives: the prompt, the sampler, the resolution, which model to use, and - if you're on the v4 line - the per-character prompt layout. You build the payload here, then wire it into NovelAI_Request, which actually makes the request.

This is the node you'll spend most of your time on, because it's the one with opinions. It's not a dumb passthrough: it remaps your inputs into NovelAI's internal parameter names and does a couple of transformations you'd otherwise have to do by hand.

How it works

Under the hood the node builds a BaseRequest, then wraps it in a NovelAITXT2IMGPayload. The remapping is where it earns its keep:

• cfg_scale becomes scale, sampler becomes sampler, n_iter becomes n_samples, and scheduler becomes noise_schedule - NovelAI's API names, not ComfyUI's.
• The selected ucPreset string gets appended to your negative_prompt automatically. There are three presets: a long quality-filter list, a shorter one, and an empty one for when you want to handle negatives yourself.
• skip_cfg_above_sigma_enable (a v4-specific trick) computes a per-resolution sigma value from a lookup table - or a fallback formula if your resolution isn't in the table - and only activates when you toggle it on.

The result is a payload object plus a payload_dict (a DICT) that shows you exactly what's about to be sent. If you're debugging, that dict is gold: it's the ground truth of what NovelAI will receive.

Inputs a beginner actually sets

Most of these are self-explanatory if you've used any diffusion UI. The ones worth knowing:

• prompt and negative_prompt - plain text, defaults are placeholders.
• model - this is where NovelAI's lineup lives: nai-diffusion-4-5-full and -curated, nai-diffusion-4-full and -curated-preview, nai-diffusion-3, nai-diffusion-furry-3, and nai-diffusion-2. Note the "preview" and "curated" suffixes - some of these are experiments, not the flagship.
• sampler - NovelAI's k_* names (k_dpmpp_2m, k_euler_ancestral, ddim_v3, …). These are NovelAI's own sampler names; don't go looking for ComfyUI's sampler list here.
• seed (default -1), steps (28), cfg_scale (6), width/height (832×1216 portrait by default).
• characterPrompts - a LIST input, normally fed by the pack's Character_Prompt_Select node. Only meaningful on the nai-diffusion-4 models; see that node's article for how it works.
• ucPreset and skip_cfg_above_sigma_enable - covered above. Both are v3/v4-era niceties; leave them alone until you know you need them.

Outputs are payload (NovelAITXT2IMGPayload, into NovelAI_Request) and payload_dict (DICT, for inspection).

Install

Same as the rest of the pack - this is one install for six nodes. Via ComfyUI Manager (search "ComfyUI_API_Request") or:

cd ComfyUI/custom_nodes
git clone https://github.com/DiaoDaiaChan/ComfyUI_API_Request

Restart ComfyUI. The only extra dependency is python-dotenv; there are no model downloads because the models live on NovelAI's servers.

Common issues

The classic beginner mistake is treating the model list like a free buffet: a "curated-preview" v4 model can behave differently from nai-diffusion-4-full on the same prompt, and character prompts (characterPrompts) simply won't do anything on nai-diffusion-3 or nai-diffusion-2. Also remember the payload only describes a generation - you haven't spent any Anlas until you connect the payload to NovelAI_Request and run. If you're iterating on settings, get the payload right first; each run of the request node is a billed generation.