Vito
cbc53ee9f1
* feat(resolver): trace multi-stage chains through pass-through nodes The inspection resolver (server and local Inspect) now compares a sampler with the earlier sampler its latent continues through image-space pass-through nodes: decode, image filter, model upscale, scale and encode, listed in a new validated `lineage` section of the semantics table. Equal prompts resolve (hires-fix, refiner); different prompts are multistage-differs. A later stage whose positive adds conditioning to the shared base prompt through an additive composite (ConditioningCombine or Concat) keeps the base pair, the text legacy linked (three-stage Flux). Image-edit batches stay ambiguous and run-time text stays untraceable. Ingestion keeps the default profile (D1). A card no longer shows a seed when the inspector names a chain stage whose seed differs from the parser's pick. * feat(resolver): leave out run-time PromptManager parts, labelled Maintainer option A: when a PromptManager prepend_text or append_text is fed directly by a node whose output is decided at run time (LoraManager's TriggerWord Toggle, now a runtime class, or PromptSearchList), the prompt resolves to the node's own text, as legacy showed. The side's source is runtime-parts-omitted and the resolved reason says that text added at run time, such as trigger words, is not included; the prompt block shows that line (D-NEW-R1). The toggle's output is never computed. Literal and traced parts are combined as before; an indirect run-time part, a run-time main text or an unknown feeder keep their outcome. The browser port changes identically. Ingestion is unchanged. * fix(resolver): keep run-time and added-conditioning labels honest A text that left out a run-time PromptManager part is now labelled when any node on its traced path omitted it, so the label survives encoders, concatenations and chained PromptManagers. Comparisons include what a text leaves out, so a partial prompt never equals a complete one: a branch with trigger words and one without stay ambiguous, and such stages are multistage-differs. Only classes listed as run time are left out. The added-conditioning floor labels its positive (D126, "Conditioning added at a later stage is not included."); the run-time line is D125. A card hides its seed only when the stage the inspector names reads a different seed than the parser; an agreeing seed is kept, as on main. The browser port moves its lineage, latent and part helpers into local-resolver-stages.mjs, keeping parity.
333 lines
12 KiB
Python
333 lines
12 KiB
Python
"""Synthetic multi-stage graphs for the pass-through lineage rule.
|
|
|
|
Each builder reproduces one graph shape with neutral texts and generic names:
|
|
a hires-fix or refiner chain (sampler, decode, image filter, model upscale,
|
|
scale, encode, sampler), a three-stage Flux run whose last stage combines the
|
|
base prompt with a style trigger, an image-edit batch of samplers with distinct
|
|
prompts, and a chain whose prompt is generated at run time. They are shared by
|
|
the resolver tests and the browser-port parity cases; no user file is involved.
|
|
"""
|
|
|
|
import copy
|
|
|
|
POSITIVE = "a lighthouse on a cliff, golden hour"
|
|
NEGATIVE = "blurry, lowres"
|
|
OTHER = "a red bicycle in the rain"
|
|
STYLE = "film grain, analog photo"
|
|
|
|
|
|
def encode(text):
|
|
return {"class_type": "CLIPTextEncode", "inputs": {"text": text, "clip": ["1", 1]}}
|
|
|
|
|
|
def prompt_manager(text, prepend=""):
|
|
return {
|
|
"class_type": "PromptManager",
|
|
"inputs": {
|
|
"text": text,
|
|
"category": "",
|
|
"tags": "",
|
|
"search_text": "",
|
|
"prepend_text": prepend,
|
|
"append_text": "",
|
|
"clip": ["1", 1],
|
|
},
|
|
}
|
|
|
|
|
|
def ksampler(positive, negative, latent, denoise=1.0):
|
|
return {
|
|
"class_type": "KSampler",
|
|
"inputs": {
|
|
"seed": 7,
|
|
"steps": 20,
|
|
"cfg": 7.0,
|
|
"sampler_name": "euler",
|
|
"scheduler": "normal",
|
|
"denoise": denoise,
|
|
"model": ["1", 0],
|
|
"positive": positive,
|
|
"negative": negative,
|
|
"latent_image": latent,
|
|
},
|
|
}
|
|
|
|
|
|
def hires_fix(second_positive=None, filter_class="Image Filter"):
|
|
"""Two KSamplers joined in image space, like a hires-fix or refiner run.
|
|
|
|
Sampler 10 runs on a custom batch latent; its output goes through a tiled
|
|
decode, an interactive image filter, a model upscale and a scale (sized by a
|
|
resolution node reading the filtered image), then is encoded again for
|
|
sampler 20 (denoise 0.25), which is decoded and saved.
|
|
"""
|
|
graph = {
|
|
"1": {"class_type": "CheckpointLoaderSimple", "inputs": {"ckpt_name": "m"}},
|
|
"2": prompt_manager(POSITIVE),
|
|
"3": encode(NEGATIVE),
|
|
"4": {
|
|
"class_type": "EmptyLatentBatch",
|
|
"inputs": {"width": 1024, "height": 512, "batch_size": 4},
|
|
},
|
|
"10": ksampler(["2", 0], ["3", 0], ["4", 0]),
|
|
"11": {
|
|
"class_type": "VAEDecodeTiled",
|
|
"inputs": {"tile_size": 512, "samples": ["10", 0], "vae": ["1", 2]},
|
|
},
|
|
"12": {
|
|
"class_type": filter_class,
|
|
"inputs": {"timeout": 600, "ontimeout": "send none", "images": ["11", 0]},
|
|
},
|
|
"13": {"class_type": "UpscaleModelLoader", "inputs": {"model_name": "x4"}},
|
|
"14": {
|
|
"class_type": "ImageUpscaleWithModel",
|
|
"inputs": {"upscale_model": ["13", 0], "image": ["12", 0]},
|
|
},
|
|
"15": {
|
|
"class_type": "ResolutionCalculator",
|
|
"inputs": {"scale_factor": 1.5, "image": ["12", 0]},
|
|
},
|
|
"16": {
|
|
"class_type": "ImageScale",
|
|
"inputs": {
|
|
"upscale_method": "nearest-exact",
|
|
"width": ["15", 0],
|
|
"height": ["15", 1],
|
|
"crop": "disabled",
|
|
"image": ["14", 0],
|
|
},
|
|
},
|
|
"17": {
|
|
"class_type": "VAEEncode",
|
|
"inputs": {"pixels": ["16", 0], "vae": ["1", 2]},
|
|
},
|
|
"20": ksampler(["2", 0], ["3", 0], ["17", 0], denoise=0.25),
|
|
"21": {
|
|
"class_type": "VAEDecodeTiled",
|
|
"inputs": {"tile_size": 512, "samples": ["20", 0], "vae": ["1", 2]},
|
|
},
|
|
"22": {
|
|
"class_type": "SaveImage",
|
|
"inputs": {"filename_prefix": "hires", "images": ["21", 0]},
|
|
},
|
|
"23": {"class_type": "PlaySound|pysssss", "inputs": {"any": ["21", 0]}},
|
|
}
|
|
if second_positive is not None:
|
|
graph["5"] = encode(second_positive)
|
|
graph["20"]["inputs"]["positive"] = ["5", 0]
|
|
return graph
|
|
|
|
|
|
_COMPOSITE_INPUTS = {
|
|
"ConditioningCombine": ("conditioning_1", "conditioning_2"),
|
|
"ConditioningConcat": ("conditioning_to", "conditioning_from"),
|
|
"ConditioningAverage": ("conditioning_to", "conditioning_from"),
|
|
}
|
|
|
|
|
|
def flux_three_stage(combined=(("30", 0), ("31", 0)), composite="ConditioningCombine"):
|
|
"""FluxGuidance base, a refiner through an image filter, then a tiled upscale.
|
|
|
|
Samplers 40 and 50 share FluxGuidance(PromptManager 30) and a zeroed empty
|
|
negative. Sampler 50 refines 40's filtered image (denoise 0.25). The final
|
|
UltimateSDUpscale 60 reads 50's image, scaled down, and a positive that is
|
|
``composite`` of ``combined`` (by default the base PromptManager and a
|
|
style-trigger encoder 31).
|
|
"""
|
|
names = _COMPOSITE_INPUTS[composite]
|
|
graph = {
|
|
"1": {"class_type": "UNETLoader", "inputs": {"unet_name": "flux"}},
|
|
"2": {"class_type": "DualCLIPLoader", "inputs": {"type": "flux"}},
|
|
"3": {"class_type": "VAELoader", "inputs": {"vae_name": "ae"}},
|
|
"30": prompt_manager(POSITIVE),
|
|
"31": encode(STYLE),
|
|
"32": {
|
|
"class_type": "CLIPTextEncode",
|
|
"inputs": {"text": "", "clip": ["2", 0]},
|
|
},
|
|
"33": {
|
|
"class_type": "ConditioningZeroOut",
|
|
"inputs": {"conditioning": ["32", 0]},
|
|
},
|
|
"34": {
|
|
"class_type": "FluxGuidance",
|
|
"inputs": {"guidance": 3.5, "conditioning": ["30", 0]},
|
|
},
|
|
"35": {
|
|
"class_type": composite,
|
|
"inputs": dict(zip(names, map(list, combined))),
|
|
},
|
|
"36": {
|
|
"class_type": "EmptyLatentBatch",
|
|
"inputs": {"width": 1080, "height": 1920, "batch_size": 4},
|
|
},
|
|
"40": ksampler(["34", 0], ["33", 0], ["36", 0]),
|
|
"41": {
|
|
"class_type": "VAEDecode",
|
|
"inputs": {"samples": ["40", 0], "vae": ["3", 0]},
|
|
},
|
|
"42": {
|
|
"class_type": "Image Filter",
|
|
"inputs": {"timeout": 600, "images": ["41", 0]},
|
|
},
|
|
"43": {"class_type": "PreviewImage", "inputs": {"images": ["42", 0]}},
|
|
"44": {
|
|
"class_type": "VAEEncode",
|
|
"inputs": {"pixels": ["42", 0], "vae": ["3", 0]},
|
|
},
|
|
"50": ksampler(["34", 0], ["33", 0], ["44", 0], denoise=0.25),
|
|
"51": {
|
|
"class_type": "VAEDecode",
|
|
"inputs": {"samples": ["50", 0], "vae": ["3", 0]},
|
|
},
|
|
"52": {
|
|
"class_type": "ImageScaleDownBy",
|
|
"inputs": {"scale_by": 0.75, "images": ["51", 0]},
|
|
},
|
|
"53": {"class_type": "PreviewImage", "inputs": {"images": ["51", 0]}},
|
|
"54": {
|
|
"class_type": "Image Comparer (rgthree)",
|
|
"inputs": {"image_a": ["51", 0], "image_b": ["42", 0]},
|
|
},
|
|
"60": {
|
|
"class_type": "UltimateSDUpscale",
|
|
"inputs": {
|
|
"upscale_by": 2.0,
|
|
"seed": 7,
|
|
"steps": 4,
|
|
"denoise": 0.35,
|
|
"image": ["52", 0],
|
|
"model": ["1", 0],
|
|
"positive": ["35", 0],
|
|
"negative": ["33", 0],
|
|
"vae": ["3", 0],
|
|
},
|
|
},
|
|
"61": {
|
|
"class_type": "SaveImage",
|
|
"inputs": {"filename_prefix": "flux", "images": ["60", 0]},
|
|
},
|
|
}
|
|
return graph
|
|
|
|
|
|
def edit_batch(count=3):
|
|
"""Image-edit samplers with distinct prompts, all saved with one prefix."""
|
|
graph = {
|
|
"1": {"class_type": "CheckpointLoaderSimple", "inputs": {"ckpt_name": "m"}},
|
|
"2": {"class_type": "LoadImage", "inputs": {"image": "ref.png"}},
|
|
"3": {
|
|
"class_type": "VAEEncode",
|
|
"inputs": {"pixels": ["2", 0], "vae": ["1", 2]},
|
|
},
|
|
"4": {"class_type": "PrimitiveString", "inputs": {"value": "edit"}},
|
|
}
|
|
for index in range(count):
|
|
base = 100 + index * 10
|
|
graph[str(base)] = {
|
|
"class_type": "TextEncodeQwenImageEdit",
|
|
"inputs": {
|
|
"prompt": f"edit number {index}: change the pose",
|
|
"clip": ["1", 1],
|
|
},
|
|
}
|
|
graph[str(base + 1)] = ksampler([str(base), 0], [str(base), 0], ["3", 0])
|
|
graph[str(base + 2)] = {
|
|
"class_type": "VAEDecode",
|
|
"inputs": {"samples": [str(base + 1), 0], "vae": ["1", 2]},
|
|
}
|
|
graph[str(base + 3)] = {
|
|
"class_type": "SaveImage",
|
|
"inputs": {"filename_prefix": ["4", 0], "images": [str(base + 2), 0]},
|
|
}
|
|
return graph
|
|
|
|
|
|
def trigger_toggle(words="trig9word"):
|
|
"""LoraManager's TriggerWord Toggle: its output is decided at run time."""
|
|
return {
|
|
"class_type": "TriggerWord Toggle (LoraManager)",
|
|
"inputs": {
|
|
"group_mode": True,
|
|
"default_active": True,
|
|
"toggle_trigger_words": {"__value__": [{"text": words, "active": True}]},
|
|
"orinalMessage": words,
|
|
"trigger_words": ["7", 2],
|
|
},
|
|
}
|
|
|
|
|
|
def hires_fix_with_trigger_words():
|
|
"""``hires_fix`` whose PromptManager prepends LoRA trigger words (node 8)."""
|
|
graph = hires_fix()
|
|
graph["7"] = {
|
|
"class_type": "Lora Loader (LoraManager)",
|
|
"inputs": {"model": ["1", 0], "clip": ["1", 1], "text": "<lora:x:1>"},
|
|
}
|
|
graph["8"] = trigger_toggle()
|
|
graph["2"]["inputs"]["prepend_text"] = ["8", 0]
|
|
return graph
|
|
|
|
|
|
def prompt_manager_text(text, prepend=""):
|
|
"""PromptManagerText: a STRING PromptManager (prepend/main/append)."""
|
|
return {
|
|
"class_type": "PromptManagerText",
|
|
"inputs": {
|
|
"text": text,
|
|
"category": "",
|
|
"tags": "",
|
|
"search_text": "",
|
|
"prepend_text": prepend,
|
|
"append_text": "",
|
|
},
|
|
}
|
|
|
|
|
|
def trigger_words_through(shape):
|
|
"""``hires_fix_with_trigger_words`` with nodes between the toggle and sampler.
|
|
|
|
``text-encode``: PromptManagerText (prepend <- toggle) -> CLIPTextEncode.
|
|
``chained``: a PromptManager whose ``text`` reads that PromptManagerText.
|
|
``concat``: PromptManagerText -> StringConcatenate(..., "extra") -> encoder.
|
|
"""
|
|
graph = hires_fix_with_trigger_words()
|
|
graph["9"] = prompt_manager_text(POSITIVE, ["8", 0])
|
|
if shape == "chained":
|
|
graph["2"]["inputs"].update(prepend_text="", text=["9", 0])
|
|
elif shape == "text-encode":
|
|
graph["2"] = {"class_type": "CLIPTextEncode", "inputs": {"text": ["9", 0]}}
|
|
else:
|
|
graph["19"] = {
|
|
"class_type": "StringConcatenate",
|
|
"inputs": {"string_a": ["9", 0], "string_b": "extra", "delimiter": ", "},
|
|
}
|
|
graph["2"] = {"class_type": "CLIPTextEncode", "inputs": {"text": ["19", 0]}}
|
|
return graph
|
|
|
|
|
|
def branches_with_and_without_trigger_words():
|
|
"""Two saved branches: one PromptManager prepends trigger words, one does not."""
|
|
graph = hires_fix_with_trigger_words()
|
|
graph["9"] = prompt_manager(POSITIVE)
|
|
graph["20"]["inputs"].update(positive=["9", 0], latent_image=["4", 0])
|
|
return graph
|
|
|
|
|
|
def generated_text_chain():
|
|
"""``hires_fix`` whose positive text comes from a run-time text generator."""
|
|
graph = hires_fix()
|
|
graph["2"] = {
|
|
"class_type": "CLIPTextEncode",
|
|
"inputs": {"text": ["6", 0], "clip": ["1", 1]},
|
|
}
|
|
graph["6"] = {
|
|
"class_type": "TextGenerate",
|
|
"inputs": {"prompt": "write a prompt", "model": ["1", 0]},
|
|
}
|
|
return graph
|
|
|
|
|
|
def copied(graph):
|
|
return copy.deepcopy(graph)
|