"""NovelAI V4+ / V5 wire payload mapping."""
from typing import Any
from .constants import Model, is_v4_model, is_v5_model, params_version_for
from .models import GenerationRequest, NovelAIGenerationPlan
# V5 replaced the integer ``ucPreset`` (0-3) with a string enum.
_UC_PRESET_IDS = ("heavy", "light", "furryFocus", "humanFocus")
def _caption(
text: str,
*,
char_captions: list[dict[str, Any]] | None = None,
) -> dict[str, Any]:
return {"caption": {"base_caption": text, "char_captions": char_captions or []}}
[docs]
def build_payload(
plan: NovelAIGenerationPlan,
*,
model: str,
) -> dict[str, Any]:
"""Map an already-resolved plan to NovelAI's V4.5/V5 request shape."""
payload: dict[str, Any] = {
"action": "generate",
"input": plan.prompt,
"model": model,
"parameters": {
"width": plan.width,
"height": plan.height,
"steps": plan.steps,
"scale": plan.scale,
"sampler": plan.sampler,
"seed": plan.seed,
"n_samples": 1,
"negative_prompt": plan.negative_prompt,
"qualityToggle": True,
"params_version": params_version_for(Model(model)),
"stream": "msgpack",
"v4_prompt": _caption(
plan.base_caption,
char_captions=list(plan.char_captions),
)
| {"use_coords": plan.use_coords, "use_order": True},
"v4_negative_prompt": _caption(plan.negative_prompt) | {"legacy_uc": False},
"characterPrompts": list(plan.character_prompts),
"add_original_image": True,
"autoSmea": False,
"cfg_rescale": 0,
"controlnet_strength": 1,
"dynamic_thresholding": False,
"image_format": "png",
"legacy": False,
"legacy_uc": False,
"legacy_v3_extend": False,
"noise_schedule": "karras",
"normalize_reference_strength_multiple": True,
"prefer_brownian": True,
"use_coords": plan.use_coords,
},
}
if is_v5_model(Model(model)):
payload["parameters"]["ucPresetId"] = "heavy"
payload["parameters"]["qualityPresetId"] = "standard"
else:
payload["parameters"]["ucPreset"] = 0
return payload
def _enum_value(value: object) -> object:
return getattr(value, "value", value)
[docs]
def build_generation_payload(request: GenerationRequest) -> dict[str, Any]:
"""Serialize every supported generation and conditioning option."""
prompt = request.effective_prompt
negative_prompt = request.effective_negative_prompt
characters = [
{
"prompt": item.prompt,
"uc": item.negative_prompt,
"center": {"x": item.x, "y": item.y},
"enabled": item.enabled,
}
for item in request.character_prompts
]
char_captions = [
{
"char_caption": item.prompt,
"centers": [{"x": item.x, "y": item.y}],
}
for item in request.character_prompts
if item.enabled
]
negative_char_captions = [
{
"char_caption": item.negative_prompt,
"centers": [{"x": item.x, "y": item.y}],
}
for item in request.character_prompts
if item.enabled and item.negative_prompt
]
parameters: dict[str, Any] = {
"width": request.width,
"height": request.height,
"n_samples": request.n_samples,
"steps": request.steps,
"scale": request.scale,
"sampler": _enum_value(request.sampler),
"seed": request.seed,
"negative_prompt": negative_prompt,
"qualityToggle": request.quality,
"params_version": params_version_for(request.model),
"dynamic_thresholding": request.dynamic_thresholding,
"cfg_rescale": request.cfg_rescale,
"noise_schedule": _enum_value(request.noise_schedule),
"add_original_image": request.add_original_image,
"controlnet_strength": request.controlnet_strength,
"characterPrompts": characters,
"skip_cfg_above_sigma": request.skip_cfg_above_sigma,
"legacy": request.legacy,
"legacy_v3_extend": request.legacy_v3_extend,
}
if is_v5_model(request.model):
# V5 uses string enums where older generations used ints/bools.
parameters["ucPresetId"] = _UC_PRESET_IDS[request.uc_preset]
parameters["qualityPresetId"] = "standard" if request.quality else "none"
if request.straight_alpha:
parameters["straight_alpha"] = True
else:
parameters["ucPreset"] = request.uc_preset
optional: dict[str, Any] = {
"extra_noise_seed": request.extra_noise_seed,
"image": request.image,
"mask": request.mask,
"strength": request.strength,
"noise": request.noise,
"controlnet_condition": request.controlnet_condition,
"controlnet_model": request.controlnet_model,
}
if not is_v5_model(request.model):
# V5 clients do not send the SMEA ``sm``/``sm_dyn`` pair — autoSmea is
# the only SMEA control (emitted in the ``is_v4_model`` block below).
optional["sm"] = request.smea
optional["sm_dyn"] = request.smea_dynamic
parameters.update({
key: value for key, value in optional.items() if value is not None
})
if request.references:
parameters["reference_image_multiple"] = list(request.references)
if request.reference_information:
parameters["reference_information_extracted_multiple"] = list(
request.reference_information
)
if request.reference_strengths:
parameters["reference_strength_multiple"] = list(
request.reference_strengths
)
if is_v4_model(request.model):
parameters.update({
"autoSmea": request.auto_smea,
"normalize_reference_strength_multiple": (
request.normalize_reference_strengths
),
"deliberate_euler_ancestral_bug": (request.deliberate_euler_ancestral_bug),
"prefer_brownian": request.prefer_brownian,
"use_coords": request.use_coords or bool(characters),
"legacy_uc": request.legacy_uc,
"stream": "msgpack",
})
if request.inpaint_img2img_strength is not None:
parameters["inpaintImg2ImgStrength"] = request.inpaint_img2img_strength
parameters["v4_prompt"] = {
"caption": {
"base_caption": request.effective_base_caption,
"char_captions": char_captions,
},
"use_coords": parameters["use_coords"],
"use_order": request.use_order,
}
parameters["v4_negative_prompt"] = {
"caption": {
"base_caption": negative_prompt,
"char_captions": negative_char_captions,
},
"legacy_uc": request.legacy_uc,
}
else:
parameters["sm"] = request.smea or False
parameters["sm_dyn"] = request.smea_dynamic or False
return {
"action": request.action.value,
"input": prompt,
"model": request.model.value,
"parameters": parameters,
}