Skip to content

The most powerful and modular diffusion model GUI, api and backend with a graph/nodes interface. The fastest local inference engine in the world.

[Partner Nodes] feat(Quiver): add Arrow 2 models with reasoning effort to the SVG nodes (#16478)

Alexander Piskun512389263395
diff --git a/comfy_api_nodes/apis/quiver.py b/comfy_api_nodes/apis/quiver.py
index bc8708754..adc7a52e0 100644
--- a/comfy_api_nodes/apis/quiver.py
+++ b/comfy_api_nodes/apis/quiver.py
@@ -6,8 +6,9 @@ class QuiverImageObject(BaseModel):
class QuiverTextToSVGRequest(BaseModel):
- model: str = Field(default="arrow-preview")
+ model: str = Field(...)
prompt: str = Field(...)
+ reasoning_effort: str | None = Field(default=None)
instructions: str | None = Field(default=None)
references: list[QuiverImageObject] | None = Field(default=None, max_length=4)
temperature: float | None = Field(default=None, ge=0, le=2)
@@ -16,8 +17,9 @@ class QuiverTextToSVGRequest(BaseModel):
class QuiverImageToSVGRequest(BaseModel):
- model: str = Field(default="arrow-preview")
+ model: str = Field(...)
image: QuiverImageObject = Field(...)
+ reasoning_effort: str | None = Field(default=None)
auto_crop: bool | None = Field(default=None)
target_size: int | None = Field(default=None, ge=128, le=4096)
temperature: float | None = Field(default=None, ge=0, le=2)
diff --git a/comfy_api_nodes/nodes_quiver.py b/comfy_api_nodes/nodes_quiver.py
index 34929fa0c..08639eacb 100644
--- a/comfy_api_nodes/nodes_quiver.py
+++ b/comfy_api_nodes/nodes_quiver.py
@@ -17,7 +17,78 @@ from comfy_api_nodes.util import (
)
from comfy_extras.nodes_images import SVG
-_ARROW_MODELS = ["arrow-1.1", "arrow-1.1-max", "arrow-preview"]
+_ARROW_MODELS = ["arrow-2", "arrow-2-telos", "arrow-1.1", "arrow-1.1-max", "arrow-preview"]
+_EFFORT_LEVELS = ["low", "medium", "high", "xhigh"]
+_TOKEN_MODEL_RATES = {"arrow-2": (4, 20), "arrow-2-telos": (6, 30)}
+_FIXED_GENERATION_USD = {"arrow-1.1": 0.286, "arrow-1.1-max": 0.3575, "arrow-preview": 0.429}
+_FIXED_VECTORIZATION_USD = {"arrow-1.1": 0.2145, "arrow-1.1-max": 0.286, "arrow-preview": 0.429}
+
+_GENERATION_TOKENS = {
+ "arrow-2": {
+ "low": (713, 1100, 723, 20000),
+ "medium": (713, 2200, 723, 34000),
+ "high": (713, 2200, 723, 36000),
+ "xhigh": (713, 2200, 723, 38000),
+ },
+ "arrow-2-telos": {
+ "low": (476, 1100, 749, 36000),
+ "medium": (476, 1300, 749, 67000),
+ "high": (476, 1300, 749, 71000),
+ "xhigh": (476, 10000, 749, 106000),
+ },
+}
+_VECTORIZATION_TOKENS = {
+ "arrow-2": {
+ "low": (591, 600, 783, 23000),
+ "medium": (591, 1300, 783, 36000),
+ "high": (591, 1300, 783, 38000),
+ "xhigh": (591, 1300, 783, 40000),
+ },
+ "arrow-2-telos": {
+ "low": (394, 300, 522, 47000),
+ "medium": (394, 300, 522, 48000),
+ "high": (394, 300, 522, 101000),
+ "xhigh": (394, 12000, 522, 115000),
+ },
+}
+
+
+def _effort_bands(model, tokens):
+ rate_in, rate_out = _TOKEN_MODEL_RATES[model]
+ effort = "widgets.reasoning_effort"
+
+ def band(level):
+ min_in, min_out, max_in, max_out = tokens[model][level]
+ return (
+ '{"type":"range_usd",'
+ f'"min_usd":({min_in} * {rate_in} + {min_out} * {rate_out}) * 1.43 / 1000000,'
+ f'"max_usd":({max_in} * {rate_in} + {max_out} * {rate_out}) * 1.43 / 1000000,'
+ '"format":{"approximate":true}}'
+ )
+
+ levels = [f'{effort} = "{level}" ? {band(level)}' for level in ("low", "medium", "xhigh")]
+ return " : ".join([*levels, band("high")])
+
+
+def _arrow_price_badge(tokens, fixed_usd):
+ branches = [f'widgets.model = "{model}" ? {{"type":"usd","usd":{usd}}}' for model, usd in fixed_usd.items()]
+ branches.append(f'widgets.model = "arrow-2-telos" ? ({_effort_bands("arrow-2-telos", tokens)})')
+ branches.append(f'({_effort_bands("arrow-2", tokens)})')
+ return IO.PriceBadge(
+ depends_on=IO.PriceBadgeDepends(widgets=["model", "reasoning_effort"]),
+ expr="(" + " : ".join(branches) + ")",
+ )
+
+
+def _reasoning_effort_input():
+ return IO.Combo.Input(
+ "reasoning_effort",
+ options=_EFFORT_LEVELS,
+ default="high",
+ optional=True,
+ tooltip="How much reasoning the model spends before drawing. Higher levels improve "
+ "detail and cost more tokens. Only used by the Arrow 2 models.",
+ )
def _arrow_sampling_inputs():
@@ -104,6 +175,7 @@ class QuiverTextToSVGNode(IO.ComfyNode):
tooltip="Seed to determine if node should re-run; "
"actual results are nondeterministic regardless of seed.",
),
+ _reasoning_effort_input(),
],
outputs=[
IO.SVG.Output(),
@@ -114,18 +186,7 @@ class QuiverTextToSVGNode(IO.ComfyNode):
IO.Hidden.unique_id,
],
is_api_node=True,
- price_badge=IO.PriceBadge(
- depends_on=IO.PriceBadgeDepends(widgets=["model"]),
- expr="""
- (
- $contains(widgets.model, "max")
- ? {"type":"usd","usd":0.3575}
- : $contains(widgets.model, "preview")
- ? {"type":"usd","usd":0.429}
- : {"type":"usd","usd":0.286}
- )
- """,
- ),
+ price_badge=_arrow_price_badge(_GENERATION_TOKENS, _FIXED_GENERATION_USD),
)
@classmethod
@@ -134,6 +195,7 @@ class QuiverTextToSVGNode(IO.ComfyNode):
prompt: str,
model: dict,
seed: int,
+ reasoning_effort: str = None,
instructions: str = None,
reference_images: IO.Autogrow.Type = None,
) -> IO.NodeOutput:
@@ -145,8 +207,6 @@ class QuiverTextToSVGNode(IO.ComfyNode):
for key in reference_images:
url = await upload_image_to_comfyapi(cls, reference_images[key], mime_type="image/png")
references.append(QuiverImageObject(url=url))
- if len(references) > 4:
- raise ValueError("Maximum 4 reference images are allowed.")
instructions_val = instructions.strip() if instructions else None
if instructions_val == "":
@@ -158,6 +218,7 @@ class QuiverTextToSVGNode(IO.ComfyNode):
response_model=QuiverSVGResponse,
data=QuiverTextToSVGRequest(
model=model["model"],
+ reasoning_effort=reasoning_effort if model["model"] in _TOKEN_MODEL_RATES else None,
prompt=prompt,
instructions=instructions_val,
references=references,
@@ -198,10 +259,11 @@ class QuiverImageToSVGNode(IO.ComfyNode):
[
IO.Int.Input(
"target_size",
- default=1024,
- min=128,
+ default=0,
+ min=0,
max=4096,
- tooltip="Square resize target in pixels.",
+ tooltip="Square resize target in pixels, 128 to 4096. 0 keeps the source "
+ "image size, which vectorizes more cleanly than forcing a resize.",
advanced=True,
),
*_arrow_sampling_inputs(),
@@ -220,6 +282,7 @@ class QuiverImageToSVGNode(IO.ComfyNode):
tooltip="Seed to determine if node should re-run; "
"actual results are nondeterministic regardless of seed.",
),
+ _reasoning_effort_input(),
],
outputs=[
IO.SVG.Output(),
@@ -230,18 +293,7 @@ class QuiverImageToSVGNode(IO.ComfyNode):
IO.Hidden.unique_id,
],
is_api_node=True,
- price_badge=IO.PriceBadge(
- depends_on=IO.PriceBadgeDepends(widgets=["model"]),
- expr="""
- (
- $contains(widgets.model, "max")
- ? {"type":"usd","usd":0.3575}
- : $contains(widgets.model, "preview")
- ? {"type":"usd","usd":0.429}
- : {"type":"usd","usd":0.286}
- )
- """,
- ),
+ price_badge=_arrow_price_badge(_VECTORIZATION_TOKENS, _FIXED_VECTORIZATION_USD),
)
@classmethod
@@ -251,6 +303,7 @@ class QuiverImageToSVGNode(IO.ComfyNode):
auto_crop: bool,
model: dict,
seed: int,
+ reasoning_effort: str = None,
) -> IO.NodeOutput:
image_url = await upload_image_to_comfyapi(cls, image, mime_type="image/png")
@@ -260,9 +313,10 @@ class QuiverImageToSVGNode(IO.ComfyNode):
response_model=QuiverSVGResponse,
data=QuiverImageToSVGRequest(
model=model["model"],
+ reasoning_effort=reasoning_effort if model["model"] in _TOKEN_MODEL_RATES else None,
image=QuiverImageObject(url=image_url),
auto_crop=auto_crop if auto_crop else None,
- target_size=model.get("target_size"),
+ target_size=model.get("target_size") or None,
temperature=model.get("temperature"),
top_p=model.get("top_p"),
presence_penalty=model.get("presence_penalty"),