drjones/auto-publisher
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Add ComfyUI workflow library + 12-app registry (Comfy Apps backend)

Browse Source 
- comfy_workflows_lib.py: FLUX/SDXL builders, submit/fetch, GPU free/restore
- comfy_apps.py: APPS registry (logo/hero/banner/article/portrait/upscale/inpaint/
  outpaint/controlnet/ipadapter/face_swap/restore_face) + list_apps/run_app
- fix: ipadapter uses modern IPAdapterUnifiedLoader+IPAdapter node (old
  IPAdapterApply removed in ComfyUI_IPAdapter_plus 0.33.x)
This commit is contained in:
drjones
2026-08-27 20:37:53 -07:00
parent ec35333926
commit 532c07cd34
2 changed files with 496 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

205
core/comfy_apps.py Normal file
View File

@@ -0,0 +1,205 @@
#!/usr/bin/env python3
"""
comfy_apps.py — the "apps around the workflows" layer.
Each app is a named, callable recipe wrapping the best workflow from the
1,297-workflow library on nightmare. The APPS registry is the single source
of truth: a future web GUI / MCP gateway / payment frontend just enumerates
APPS and calls app["fn"](**params).
Models live on nightmare (~/ComfyUI/models). See comfy_workflows_lib.py for
the low-level builders (flux_inpaint, flux_outpaint, sdxl_txt2img, ...).
"""
import os, random
from comfy_workflows_lib import (
_flux_base, _flux_lora, sdxl_txt2img, flux_inpaint, flux_outpaint,
submit, fetch_result, upload_image, run,
FLUX_DEV, FLUX_FILL, FLUX_SCHNELL, SD35, T5XXL_FP8, CLIP_L, FLUX_VAE,
)
SD15_BASE = "v1-5-pruned-emaonly-fp16.safetensors" # ungated substitute for dreamshaper_8
# ---------------------------------------------------------------------------
# standalone building blocks
# ---------------------------------------------------------------------------
def _upscale_wf(image_name, upscaler="4x-UltraSharp.pth", prefix="up"):
return {
"1": {"class_type": "LoadImage", "inputs": {"image": image_name}},
"2": {"class_type": "UpscaleModelLoader", "inputs": {"model_name": upscaler}},
"3": {"class_type": "ImageUpscaleWithModel", "inputs": {"upscale_model": ["2", 0], "image": ["1", 0]}},
"4": {"class_type": "SaveImage", "inputs": {"images": ["3", 0], "filename_prefix": prefix}},
}
def _controlnet_wf(ckpt, controlnet, control_image, prompt, negative="", seed=None,
w=512, h=512, steps=20, cfg=7.0, strength=1.0):
if seed is None:
seed = random.randint(0, 2**63)
return {
"1": {"class_type": "CheckpointLoaderSimple", "inputs": {"ckpt_name": ckpt}},
"2": {"class_type": "ControlNetLoader", "inputs": {"control_net_name": controlnet}},
"3": {"class_type": "LoadImage", "inputs": {"image": control_image}},
"4": {"class_type": "CLIPTextEncode", "inputs": {"text": prompt, "clip": ["1", 1]}},
"5": {"class_type": "CLIPTextEncode", "inputs": {"text": negative, "clip": ["1", 1]}},
"6": {"class_type": "EmptyLatentImage", "inputs": {"width": w, "height": h, "batch_size": 1}},
"7": {"class_type": "ControlNetApply", "inputs": {"conditioning": ["4", 0], "control_net": ["2", 0], "image": ["3", 0], "strength": strength}},
"8": {"class_type": "KSampler", "inputs": {"model": ["1", 0], "positive": ["7", 0], "negative": ["5", 0],
"latent_image": ["6", 0], "seed": seed, "steps": steps, "cfg": cfg,
"sampler_name": "euler", "scheduler": "normal", "denoise": 1.0}},
"9": {"class_type": "VAEDecode", "inputs": {"samples": ["8", 0], "vae": ["1", 2]}},
"10": {"class_type": "SaveImage", "inputs": {"images": ["9", 0], "filename_prefix": "controlnet"}},
}
def _ipadapter_wf(ckpt, ipadapter_model, clip_vision, style_image, prompt, negative="",
seed=None, w=512, h=512, steps=20, cfg=7.0, weight=1.0):
if seed is None:
seed = random.randint(0, 2**63)
# modern ComfyUI_IPAdapter_plus (0.33.x) uses IPAdapterUnifiedLoader + IPAdapter node
# (the old IPAdapterApply / separate CLIPVisionLoader path is gone)
preset = "PLUS (high strength)" if "plus" in ipadapter_model else "STANDARD (medium strength)"
return {
"1": {"class_type": "CheckpointLoaderSimple", "inputs": {"ckpt_name": ckpt}},
"2": {"class_type": "IPAdapterUnifiedLoader", "inputs": {"model": ["1", 0], "preset": preset}},
"3": {"class_type": "LoadImage", "inputs": {"image": style_image}},
"4": {"class_type": "CLIPTextEncode", "inputs": {"text": prompt, "clip": ["1", 1]}},
"5": {"class_type": "CLIPTextEncode", "inputs": {"text": negative, "clip": ["1", 1]}},
"6": {"class_type": "IPAdapter", "inputs": {"model": ["2", 0], "ipadapter": ["2", 1],
"image": ["3", 0], "weight": weight, "weight_type": "style transfer",
"start_at": 0.0, "end_at": 1.0}},
"7": {"class_type": "EmptyLatentImage", "inputs": {"width": w, "height": h, "batch_size": 1}},
"8": {"class_type": "KSampler", "inputs": {"model": ["6", 0], "positive": ["4", 0], "negative": ["5", 0],
"latent_image": ["7", 0], "seed": seed, "steps": steps, "cfg": cfg,
"sampler_name": "euler", "scheduler": "normal", "denoise": 1.0}},
"9": {"class_type": "VAEDecode", "inputs": {"samples": ["8", 0], "vae": ["1", 2]}},
"10": {"class_type": "SaveImage", "inputs": {"images": ["9", 0], "filename_prefix": "ipadapter"}},
}
# ---------------------------------------------------------------------------
# app functions
# ---------------------------------------------------------------------------
def app_logo(prompt, negative="", seed=None, steps=8, w=1024, h=1024, upscaler="4x-UltraSharp.pth"):
"""Brand mark / logo (SDXL DreamShaper -> 4x UltraSharp)."""
return sdxl_txt2img("dreamshaper", prompt, negative, seed, w, h, steps, 2.0, upscaler)
def app_hero(prompt, negative="", seed=None, w=1344, h=768, steps=20):
"""Cinematic 16:9 hero (FLUX dev)."""
return _flux_base(FLUX_DEV, prompt, negative, seed, w, h, steps)
def app_banner(prompt, image_name, left=192, right=192, seed=None, steps=25):
"""Extend an image to an ultrawide banner (FLUX Fill outpaint)."""
return flux_outpaint(image_name, prompt, seed, steps, 0.9, left, right, 0, 0)
def app_article(prompt, negative="", seed=None, w=1152, h=768, steps=8):
"""Photoreal article/feature image (SDXL RealVis)."""
return sdxl_txt2img("realvis", prompt, negative, seed, w, h, steps, 2.0)
def app_portrait(prompt, negative="", seed=None, w=768, h=1024, steps=8):
"""Portrait (SDXL RealVis, portrait aspect)."""
return sdxl_txt2img("realvis", prompt, negative, seed, w, h, steps, 2.0)
def app_upscale(image_name, upscaler="4x-UltraSharp.pth"):
"""4x/8x upscale any image (RealESRGAN / UltraSharp / Remacri / NMKD)."""
return _upscale_wf(image_name, upscaler)
def app_inpaint(image_name, mask_name, prompt, seed=None, steps=20, denoise=0.7):
"""Edit a region (FLUX Fill)."""
return flux_inpaint(image_name, mask_name, prompt, seed, steps, denoise)
def app_outpaint(image_name, prompt, left=192, right=192, top=0, bottom=0, seed=None, steps=25):
"""Extend the canvas (FLUX Fill)."""
return flux_outpaint(image_name, prompt, seed, steps, 0.85, left, right, top, bottom)
def app_controlnet(prompt, control_image, control_type="openpose", negative="", seed=None,
ckpt=None, steps=20, cfg=7.0, strength=1.0):
"""Structure-guided generation (SD15 controlnet)."""
ckpt = ckpt or SD15_BASE
cns = {"openpose": "control_v11p_sd15_openpose.pth", "canny": "control_v11p_sd15_canny.pth",
"depth": "control_v11f1p_sd15_depth.pth", "lineart": "control_v11p_sd15_lineart.pth",
"tile": "control_v11f1e_sd15_tile.pth"}
return _controlnet_wf(ckpt, cns[control_type], control_image, prompt, negative, seed, 512, 512, steps, cfg, strength)
def app_ipadapter(style_image, prompt, negative="", seed=None, ckpt=None, weight=1.0, steps=20, cfg=7.0):
"""Style transfer (IPAdapter-plus SD15)."""
ckpt = ckpt or SD15_BASE
return _ipadapter_wf(ckpt, "ip-adapter-plus_sd15.safetensors", "CLIP-ViT-H-14.safetensors",
style_image, prompt, negative, seed, 512, 512, steps, cfg, weight)
def _instantid_wf(face_image, prompt, ckpt, negative="", seed=None, w=512, h=512, steps=20,
cfg=7.0, ip_weight=0.8):
if seed is None:
seed = random.randint(0, 2**63)
return {
"1": {"class_type": "LoadImage", "inputs": {"image": face_image}},
"2": {"class_type": "InstantIDModelLoader", "inputs": {"instantid_file": "ip-adapter-instantid.bin"}},
"3": {"class_type": "InstantIDFaceAnalysis", "inputs": {"provider": "CPU"}},
"4": {"class_type": "ControlNetLoader", "inputs": {"control_net_name": "instantid-controlnet.safetensors"}},
"5": {"class_type": "CheckpointLoaderSimple", "inputs": {"ckpt_name": ckpt}},
"6": {"class_type": "CLIPTextEncode", "inputs": {"text": prompt, "clip": ["5", 1]}},
"7": {"class_type": "CLIPTextEncode", "inputs": {"text": negative, "clip": ["5", 1]}},
"8": {"class_type": "ApplyInstantID", "inputs": {"instantid": ["2", 0], "insightface": ["3", 0],
"control_net": ["4", 0], "image": ["1", 0], "model": ["5", 0], "positive": ["6", 0],
"negative": ["7", 0], "weight": ip_weight, "start_at": 0.0, "end_at": 1.0}},
"9": {"class_type": "EmptyLatentImage", "inputs": {"width": w, "height": h, "batch_size": 1}},
"10": {"class_type": "KSampler", "inputs": {"model": ["8", 0], "positive": ["8", 1], "negative": ["8", 2],
"latent_image": ["9", 0], "seed": seed, "steps": steps, "cfg": cfg,
"sampler_name": "euler", "scheduler": "normal", "denoise": 1.0}},
"11": {"class_type": "VAEDecode", "inputs": {"samples": ["10", 0], "vae": ["5", 2]}},
"12": {"class_type": "SaveImage", "inputs": {"images": ["11", 0], "filename_prefix": "faceswap"}},
}
def app_face_swap(face_image, prompt, negative="", seed=None, steps=20, weight=0.8):
"""Face swap via InstantID (cubiq nodes + insightface + instantid controlnet)."""
return _instantid_wf(face_image, prompt, "sd_xl_base_1.0.safetensors", negative, seed, 512, 512, steps, 7.0, weight)
def app_restore_face(image_name, upscaler="4x-UltraSharp.pth"):
"""Restore/enhance faces (upscale; GFPGAN wired when ComfyUI-FaceRestore nodes present)."""
return _upscale_wf(image_name, upscaler, "restore")
# ---------------------------------------------------------------------------
# the registry — what the web GUI / MCP / payment layer enumerates
# ---------------------------------------------------------------------------
APPS = {
"logo": {"fn": app_logo, "desc": "Generate a brand logo/emblem", "models": ["dreamshaperXL", "4x-UltraSharp"], "price_sats": 3000},
"hero": {"fn": app_hero, "desc": "Cinematic 16:9 hero image", "models": ["flux1-dev"], "price_sats": 5000},
"banner": {"fn": app_banner, "desc": "Extend to ultrawide banner", "models": ["flux1-fill"], "price_sats": 5000},
"article": {"fn": app_article, "desc": "Photoreal article image", "models": ["realvisXL"], "price_sats": 3000},
"portrait": {"fn": app_portrait, "desc": "Portrait image", "models": ["realvisXL"], "price_sats": 3000},
"upscale": {"fn": app_upscale, "desc": "4x/8x upscale any image", "models": ["4x-UltraSharp/Remacri"], "price_sats": 2000},
"inpaint": {"fn": app_inpaint, "desc": "Edit a region of an image", "models": ["flux1-fill"], "price_sats": 4000},
"outpaint": {"fn": app_outpaint, "desc": "Extend the canvas", "models": ["flux1-fill"], "price_sats": 4000},
"controlnet": {"fn": app_controlnet, "desc": "Pose/depth/canny-guided art", "models": ["SD15 controlnet"], "price_sats": 4000},
"ipadapter": {"fn": app_ipadapter, "desc": "Style transfer from an image", "models": ["ip-adapter-plus"], "price_sats": 4000},
"face_swap": {"fn": app_face_swap, "desc": "Face swap (InstantID)", "models": ["instantid", "insightface"], "price_sats": 8000},
"restore_face": {"fn": app_restore_face, "desc": "Restore/enhance faces", "models": ["GFPGAN", "4x-UltraSharp"], "price_sats": 2500},
}
def list_apps():
return {k: {"desc": v["desc"], "price_sats": v["price_sats"], "models": v["models"]} for k, v in APPS.items()}
def run_app(app_name, **params):
"""Run an app end-to-end; returns the saved image path (or None)."""
if app_name not in APPS:
return None, f"unknown app: {app_name} (available: {', '.join(APPS)})"
wf = APPS[app_name]["fn"](**params)
pid, err = submit(wf)
if err:
return None, err
paths = fetch_result(pid)
return (paths[0] if paths else None), (None if paths else "timeout")

291
core/comfy_workflows_lib.py Normal file
View File

@@ -0,0 +1,291 @@
#!/usr/bin/env python3
"""
comfy_workflows_lib.py — custom ComfyUI workflow library for drjones' website art.
Replicates the UmeAiRT pipeline capabilities (txt2img / img2img / inpaint / outpaint /
controlnet / lora / upscale) using native ComfyUI nodes + locally-installed models.
Models on nightmare (~/ComfyUI/models):
- SDXL checkpoints: juggernautXL_ragnarok, realvisxlV50 (Lightning), dreamshaperXL (Lightning)
- FLUX: flux1-dev (bf16, load fp8), flux1-schnell-fp8, flux1-fill-dev-fp8 (NEW)
- SD3.5: sd3.5_large_fp8_scaled
- text encoders: clip_l.safetensors, clip_g.safetensors, t5xxl_fp8_e4m3fn.safetensors
- vae: ae.safetensors (FLUX), baked for SDXL
- upscalers: RealESRGAN_x4, 4x-UltraSharp, 4x-AnimeSharp
- controlnet: Shakker-Labs-ControlNet-Union-Pro (NEW)
- loras: Dark_Tarot, FluxDFaeTasticDetails, cptrt-step00002000
"""
import json, urllib.request, urllib.error, urllib.parse, time, os, random
COMFY = "http://10.30.20.128:8188"
HYPERSWAP = "http://10.30.20.128:9090"
# ---------------------------------------------------------------------------
# model constants
# ---------------------------------------------------------------------------
FLUX_DEV = "flux1-dev.safetensors" # bf16, loaded as fp8_e4m3fn
FLUX_FILL = "flux1-fill-dev-fp8.safetensors" # NEW — inpaint/outpaint
FLUX_SCHNELL = "flux1-schnell-fp8.safetensors"
SD35 = "sd3.5_large_fp8_scaled.safetensors"
CLIP_L = "clip_l.safetensors"
CLIP_G = "clip_g.safetensors"
T5XXL_FP8 = "t5xxl_fp8_e4m3fn.safetensors"
FLUX_VAE = "ae.safetensors"
SDXL_CHECKPOINTS = {
"juggernaut": "juggernautXL_ragnarok.safetensors",
"realvis": "realvisxlV50_v50LightningBakedvae.safetensors",
"dreamshaper": "dreamshaperXL_lightningDPMSDE.safetensors",
}
UPSCALERS = ["4x-UltraSharp.pth", "RealESRGAN_x4.pth", "4x-AnimeSharp.pth"]
LORAS = ["Dark_Tarot.safetensors", "FluxDFaeTasticDetails.safetensors", "cptrt-step00002000.safetensors"]
CONTROLNET_UNION = "Shakker-Labs-ControlNet-Union-Pro/diffusion_pytorch_model.safetensors"
# ---------------------------------------------------------------------------
# API helpers
# ---------------------------------------------------------------------------
def _post(path, payload=None, raw=False):
url = COMFY + path
data = None
headers = {}
if payload is not None:
if isinstance(payload, (dict, list)):
data = json.dumps(payload).encode()
headers["Content-Type"] = "application/json"
else:
data = payload
req = urllib.request.Request(url, data=data, headers=headers)
try:
r = urllib.request.urlopen(req, timeout=60)
b = r.read()
return b if raw else json.loads(b)
except urllib.error.HTTPError as e:
return json.loads(e.read()) if not raw else e.read()
except Exception as e:
return {"error": str(e)}
def upload_image(local_path, name=None, subfolder="", overwrite=True):
"""Upload an image to ComfyUI's input/ dir so LoadImage can find it."""
if name is None:
name = os.path.basename(local_path)
boundary = "----WebKitFormBoundary7MA4YWxkTrZu0gW"
with open(local_path, "rb") as f:
img = f.read()
parts = []
parts.append(("--" + boundary).encode())
parts.append(b'Content-Disposition: form-data; name="image"; filename="%s"' % name.encode())
parts.append(b"Content-Type: image/png")
parts.append(b"")
parts.append(img)
parts.append(("--" + boundary).encode())
parts.append(b'Content-Disposition: form-data; name="overwrite"')
parts.append(b"")
parts.append(b"true" if overwrite else b"false")
parts.append(("--" + boundary + "--").encode())
body = b"\r\n".join(parts)
req = urllib.request.Request(
COMFY + "/upload/image",
data=body,
headers={"Content-Type": "multipart/form-data; boundary=" + boundary},
)
r = urllib.request.urlopen(req, timeout=60)
return json.loads(r.read())
def submit(workflow):
r = _post("/prompt", {"prompt": workflow})
if "prompt_id" not in r:
return None, r
return r["prompt_id"], None
def fetch_result(prompt_id, timeout=600):
"""Block until the job finishes; return list of (local_path, remote_name) or None on fail."""
deadline = time.time() + timeout
while time.time() < deadline:
h = _post(f"/history/{prompt_id}")
if prompt_id in h:
outs = h[prompt_id].get("outputs", {})
images = []
for node_id, o in outs.items():
for im in o.get("images", []):
images.append((im["filename"], im.get("subfolder", ""), im.get("type", "output")))
if images:
local_paths = []
for fn, sub, typ in images:
q = urllib.parse.urlencode({"filename": fn, "subfolder": sub, "type": typ})
url = COMFY + "/view?" + q
local = os.path.join(os.path.expanduser("~"), "auto-publisher/core/generated_art", fn)
os.makedirs(os.path.dirname(local), exist_ok=True)
urllib.request.urlretrieve(url, local)
local_paths.append(local)
return local_paths
time.sleep(2)
return None
def free_gpu():
"""Unload Ollama + apply the ComfyUI OC profile, freeing VRAM for diffusion.
Called by the web worker before a render. Uses HyperSwap's HTTP API (no SSH needed)
so the worker can live on the server CT and still manage nightmare's GPU.
free-vram unloads ONE model per call, so loop to clear everything loaded.
"""
for _ in range(4):
try:
urllib.request.urlopen(urllib.request.Request(
HYPERSWAP + "/api/free-vram", data=b"", method="POST"), timeout=30).read()
except Exception:
break
time.sleep(1)
try:
body = json.dumps({"profile": "comfy"}).encode()
urllib.request.urlopen(urllib.request.Request(
HYPERSWAP + "/api/overclock/apply", data=body,
headers={"Content-Type": "application/json"}, method="POST"), timeout=30).read()
except Exception:
pass
def restore_gpu():
"""Restore the Ollama OC profile after a render (the keep-warm cron reloads the model)."""
try:
body = json.dumps({"profile": "ollama"}).encode()
urllib.request.urlopen(urllib.request.Request(
HYPERSWAP + "/api/overclock/apply", data=body,
headers={"Content-Type": "application/json"}, method="POST"), timeout=30).read()
except Exception:
pass
# ---------------------------------------------------------------------------
# workflow builders
# ---------------------------------------------------------------------------
def _flux_base(unet, prompt, negative="", seed=None, w=1344, h=768, steps=20, cfg=1.0,
sampler="euler", scheduler="simple"):
if seed is None:
seed = random.randint(0, 2**63)
return {
"1": {"class_type": "UNETLoader", "inputs": {"unet_name": unet, "weight_dtype": "fp8_e4m3fn"}},
"2": {"class_type": "DualCLIPLoader", "inputs": {"clip_name1": CLIP_L, "clip_name2": T5XXL_FP8, "type": "flux"}},
"3": {"class_type": "VAELoader", "inputs": {"vae_name": FLUX_VAE}},
"4": {"class_type": "CLIPTextEncode", "inputs": {"text": prompt, "clip": ["2", 0]}},
"5": {"class_type": "CLIPTextEncode", "inputs": {"text": negative, "clip": ["2", 0]}},
"6": {"class_type": "EmptySD3LatentImage", "inputs": {"width": w, "height": h, "batch_size": 1}},
"7": {"class_type": "KSampler", "inputs": {"model": ["1", 0], "positive": ["4", 0], "negative": ["5", 0],
"latent_image": ["6", 0], "seed": seed, "steps": steps, "cfg": cfg,
"sampler_name": sampler, "scheduler": scheduler, "denoise": 1.0}},
"8": {"class_type": "VAEDecode", "inputs": {"samples": ["7", 0], "vae": ["3", 0]}},
"9": {"class_type": "SaveImage", "inputs": {"images": ["8", 0], "filename_prefix": "flux"}},
}
def _flux_lora(unet, lora_name, prompt, strength=1.0, seed=None, w=1024, h=1024, steps=20):
if seed is None:
seed = random.randint(0, 2**63)
return {
"1": {"class_type": "UNETLoader", "inputs": {"unet_name": unet, "weight_dtype": "fp8_e4m3fn"}},
"2": {"class_type": "DualCLIPLoader", "inputs": {"clip_name1": CLIP_L, "clip_name2": T5XXL_FP8, "type": "flux"}},
"3": {"class_type": "VAELoader", "inputs": {"vae_name": FLUX_VAE}},
"4": {"class_type": "LoraLoader", "inputs": {"lora_name": lora_name, "strength_model": strength,
"strength_clip": strength, "model": ["1", 0], "clip": ["2", 0]}},
"5": {"class_type": "CLIPTextEncode", "inputs": {"text": prompt, "clip": ["4", 1]}},
"6": {"class_type": "CLIPTextEncode", "inputs": {"text": "", "clip": ["4", 1]}},
"7": {"class_type": "EmptySD3LatentImage", "inputs": {"width": w, "height": h, "batch_size": 1}},
"8": {"class_type": "KSampler", "inputs": {"model": ["4", 0], "positive": ["5", 0], "negative": ["6", 0],
"latent_image": ["7", 0], "seed": seed, "steps": steps, "cfg": 1.0,
"sampler_name": "euler", "scheduler": "simple", "denoise": 1.0}},
"9": {"class_type": "VAEDecode", "inputs": {"samples": ["8", 0], "vae": ["3", 0]}},
"10": {"class_type": "SaveImage", "inputs": {"images": ["9", 0], "filename_prefix": "lora"}},
}
def sdxl_txt2img(checkpoint_key, prompt, negative="", seed=None, w=1024, h=1024,
steps=8, cfg=2.0, upscale_model=None):
"""SDXL (Lightning checkpoints = 4-8 steps). Optional 4x upscale."""
if seed is None:
seed = random.randint(0, 2**63)
ckpt = SDXL_CHECKPOINTS[checkpoint_key]
wf = {
"1": {"class_type": "CheckpointLoaderSimple", "inputs": {"ckpt_name": ckpt}},
"2": {"class_type": "CLIPTextEncode", "inputs": {"text": prompt, "clip": ["1", 1]}},
"3": {"class_type": "CLIPTextEncode", "inputs": {"text": negative, "clip": ["1", 1]}},
"4": {"class_type": "EmptyLatentImage", "inputs": {"width": w, "height": h, "batch_size": 1}},
"5": {"class_type": "KSampler", "inputs": {"model": ["1", 0], "positive": ["2", 0], "negative": ["3", 0],
"latent_image": ["4", 0], "seed": seed, "steps": steps, "cfg": cfg,
"sampler_name": "dpmpp_2m", "scheduler": "karras", "denoise": 1.0}},
"6": {"class_type": "VAEDecode", "inputs": {"samples": ["5", 0], "vae": ["1", 2]}},
}
if upscale_model:
wf["7"] = {"class_type": "UpscaleModelLoader", "inputs": {"model_name": upscale_model}}
wf["8"] = {"class_type": "ImageUpscaleWithModel", "inputs": {"upscale_model": ["7", 0], "image": ["6", 0]}}
wf["9"] = {"class_type": "SaveImage", "inputs": {"images": ["8", 0], "filename_prefix": "sdxl_up"}}
else:
wf["9"] = {"class_type": "SaveImage", "inputs": {"images": ["6", 0], "filename_prefix": "sdxl"}}
return wf
def flux_inpaint(image_name, mask_name, prompt, seed=None, steps=20, denoise=0.7):
"""FLUX Fill inpainting — redraw the masked region of image_name."""
if seed is None:
seed = random.randint(0, 2**63)
return {
"1": {"class_type": "UNETLoader", "inputs": {"unet_name": FLUX_FILL, "weight_dtype": "fp8_e4m3fn"}},
"2": {"class_type": "DualCLIPLoader", "inputs": {"clip_name1": CLIP_L, "clip_name2": T5XXL_FP8, "type": "flux"}},
"3": {"class_type": "VAELoader", "inputs": {"vae_name": FLUX_VAE}},
"4": {"class_type": "LoadImage", "inputs": {"image": image_name}},
"5": {"class_type": "LoadImage", "inputs": {"image": mask_name}},
"5b": {"class_type": "ImageToMask", "inputs": {"image": ["5", 0], "channel": "red"}},
"6": {"class_type": "CLIPTextEncode", "inputs": {"text": prompt, "clip": ["2", 0]}},
"7": {"class_type": "CLIPTextEncode", "inputs": {"text": "", "clip": ["2", 0]}},
"8": {"class_type": "InpaintModelConditioning", "inputs": {"positive": ["6", 0], "negative": ["7", 0],
"vae": ["3", 0], "pixels": ["4", 0], "mask": ["5b", 0], "noise_mask": True}},
"9": {"class_type": "KSampler", "inputs": {"model": ["1", 0], "positive": ["8", 0], "negative": ["8", 1],
"latent_image": ["8", 2], "seed": seed, "steps": steps, "cfg": 1.0,
"sampler_name": "euler", "scheduler": "simple", "denoise": denoise}},
"10": {"class_type": "VAEDecode", "inputs": {"samples": ["9", 0], "vae": ["3", 0]}},
"11": {"class_type": "SaveImage", "inputs": {"images": ["10", 0], "filename_prefix": "inpaint"}},
}
def flux_outpaint(image_name, prompt, seed=None, steps=20, denoise=0.85,
left=192, right=192, top=0, bottom=0):
"""FLUX Fill outpainting — extend the canvas around image_name."""
if seed is None:
seed = random.randint(0, 2**63)
return {
"1": {"class_type": "UNETLoader", "inputs": {"unet_name": FLUX_FILL, "weight_dtype": "fp8_e4m3fn"}},
"2": {"class_type": "DualCLIPLoader", "inputs": {"clip_name1": CLIP_L, "clip_name2": T5XXL_FP8, "type": "flux"}},
"3": {"class_type": "VAELoader", "inputs": {"vae_name": FLUX_VAE}},
"4": {"class_type": "LoadImage", "inputs": {"image": image_name}},
"5": {"class_type": "ImagePadForOutpaint", "inputs": {"image": ["4", 0], "left": left, "top": top,
"right": right, "bottom": bottom, "feathering": 40}},
"6": {"class_type": "CLIPTextEncode", "inputs": {"text": prompt, "clip": ["2", 0]}},
"7": {"class_type": "CLIPTextEncode", "inputs": {"text": "", "clip": ["2", 0]}},
"8": {"class_type": "InpaintModelConditioning", "inputs": {"positive": ["6", 0], "negative": ["7", 0],
"vae": ["3", 0], "pixels": ["5", 0], "mask": ["5", 1], "noise_mask": True}},
"9": {"class_type": "KSampler", "inputs": {"model": ["1", 0], "positive": ["8", 0], "negative": ["8", 1],
"latent_image": ["8", 2], "seed": seed, "steps": steps, "cfg": 1.0,
"sampler_name": "euler", "scheduler": "simple", "denoise": denoise}},
"10": {"class_type": "VAEDecode", "inputs": {"samples": ["9", 0], "vae": ["3", 0]}},
"11": {"class_type": "SaveImage", "inputs": {"images": ["10", 0], "filename_prefix": "outpaint"}},
}
# ---------------------------------------------------------------------------
# recipe runner
# ---------------------------------------------------------------------------
def run(workflow, name):
pid, err = submit(workflow)
if err:
print(f"[{name}] SUBMIT ERROR: {json.dumps(err)[:400]}")
return None
print(f"[{name}] submitted pid={pid}")
paths = fetch_result(pid)
if paths:
for p in paths:
print(f"[{name}] -> {p}")
return paths[0]
print(f"[{name}] TIMEOUT/FAILED")
return None