File size: 15,332 Bytes
3029c38 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 | import os
import torch
from shared.utils.hf import build_hf_url
from .prompt_enhancer import IDEOGRAM4_PROMPT_ENHANCER, IDEOGRAM4_PROMPT_INFOS
_PROJECT_REPO = "DeepBeepMeep/Ideogram4"
_PROJECT_FOLDER = "ideogram4"
_TEXT_ENCODER_FOLDER = "Qwen3-VL-8B-Instruct"
_VAE_REPO = "DeepBeepMeep/Flux2"
_VAE_FILENAME = "flux2_vae.safetensors"
_TURBOTIME_MODEL_TYPE = "ideogram4_turbotime"
_PRESET_CHOICES = [
("Quality 48", "V4_QUALITY_48"),
("Default 20", "V4_DEFAULT_20"),
("Turbo 12", "V4_TURBO_12"),
]
_DEFAULT_PRESET = "V4_DEFAULT_20"
_DEFAULT_SAMPLE_SOLVER = "euler"
_DEFAULT_FLOW_SHIFT = 1.0
_SAMPLE_SOLVERS = [
("Euler", "euler"),
("RES 2M", "res_2m"),
("RES 2S", "res_2s"),
]
_SAMPLE_SOLVER_IDS = {value for _, value in _SAMPLE_SOLVERS}
_CUSTOM_SETTINGS = [
{"id": "ideogram_mu", "label": "Scheduler Mu", "name": "Scheduler Mu", "type": "float", "default": 0.0, "min": -10.0, "max": 10.0, "inc": 0.05},
{"id": "ideogram_std", "label": "Scheduler Std", "name": "Scheduler Std", "type": "float", "default": 1.75, "min": 0.1, "max": 5.0, "inc": 0.05},
]
IDEOGRAM4_INFOS = """## Ideogram 4
Ideogram 4 is an image model specialized for prompt-following, text rendering, layout, logos, posters, typography, and graphic design.
It can take a plain text prompt, but it is designed to work especially well with structured JSON prompts. In WanGP, Magic Prompt can convert a normal text prompt into Ideogram's JSON-style caption format. For best prompt adherence, use the JSON format when possible.
## Guidance phases
Ideogram 4 uses WanGP guidance phases for CFG switching.
Phase 1 uses Guidance during the high-noise part of denoising. Phase 2 uses Guidance2 after the first switch threshold. If Three Phases is enabled, phase 3 uses Guidance3 after the second switch threshold.
The official presets use two phases: most steps at CFG 7, then final low-noise polish at CFG 3.
## Scheduler custom settings
Scheduler Mu and Scheduler Std control the Ideogram logit-normal sampling schedule.
Mu shifts where denoising samples concentrate along the noise schedule. Std controls how wide that distribution is.
The official presets set these values for you. Change them mainly when matching a Comfy workflow or experimenting with a different schedule.
## LoRA multipliers
Ideogram 4 has two parallel transformers: the conditional transformer and the unconditional transformer.
A plain LoRA multiplier applies to both branches.
To set different values per branch, use ':' as conditional:unconditional. For example, 0.9:0.4 applies 0.9 to the conditional transformer and 0.4 to the unconditional transformer.
Guidance phases still use ';'. For example, 1:1.2;0.8:1.0 means phase 1 cond/uncond = 1/1.2 and phase 2 cond/uncond = 0.8/1.0."""
IDEOGRAM4_TURBOTIME_INFOS = """## Ideogram 4 TurboTime
Ideogram 4 TurboTime uses the Ideogram 4 conditional transformer with the TurboTime LoRA applied. It is designed for a few denoising steps with no CFG and no unconditional transformer branch.
LoRA multipliers apply to the conditional transformer only."""
_PRESET_SETTINGS = {
"V4_QUALITY_48": {"num_inference_steps": 48, "mu": 0.0, "std": 1.5, "switch_threshold": 184},
"V4_DEFAULT_20": {"num_inference_steps": 20, "mu": 0.0, "std": 1.75, "switch_threshold": 211},
"V4_TURBO_12": {"num_inference_steps": 12, "mu": 0.5, "std": 1.75, "switch_threshold": 302},
}
_LEGACY_CFG_OVERRIDE_KEYS = (
"ideogram_cfg_override_enabled",
"ideogram_cfg_override_start_percent",
"ideogram_cfg_override_end_percent",
)
_TURBOTIME_SETTINGS = {
"num_inference_steps": 8,
"guidance_phases": 0,
"guidance_scale": 0,
"guidance2_scale": 0,
"guidance3_scale": 0,
"switch_threshold": 0,
"switch_threshold2": 0,
"sample_solver": _DEFAULT_SAMPLE_SOLVER,
"flow_shift": _DEFAULT_FLOW_SHIFT,
}
_TURBOTIME_CUSTOM_SETTINGS = {
"ideogram_mu": 0.5,
"ideogram_std": 1.75,
}
def _time_snr_shift(shift, sigma):
return shift * sigma / (1.0 + (shift - 1.0) * sigma)
def _percent_to_switch_threshold(percent, shift=1.0):
return int(round(_time_snr_shift(float(shift), 1.0 - float(percent)) * 1000.0))
def _drop_legacy_cfg_override_settings(custom_settings):
for key in _LEGACY_CFG_OVERRIDE_KEYS:
custom_settings.pop(key, None)
def _apply_turbotime_settings(ui_defaults, preserve_user_values=False):
custom_settings = ui_defaults.get("custom_settings", None)
if not isinstance(custom_settings, dict):
custom_settings = {}
_drop_legacy_cfg_override_settings(custom_settings)
for key, value in _TURBOTIME_CUSTOM_SETTINGS.items():
if preserve_user_values:
custom_settings.setdefault(key, value)
else:
custom_settings[key] = value
for key, value in _TURBOTIME_SETTINGS.items():
if preserve_user_values and key not in {"guidance_phases", "guidance_scale", "guidance2_scale", "guidance3_scale", "switch_threshold", "switch_threshold2"}:
ui_defaults.setdefault(key, value)
else:
ui_defaults[key] = value
ui_defaults["custom_settings"] = custom_settings
ui_defaults["model_mode"] = None
def _apply_preset_settings(ui_defaults, preset_name):
preset = _PRESET_SETTINGS[preset_name]
custom_settings = ui_defaults.get("custom_settings", None)
if not isinstance(custom_settings, dict):
custom_settings = {}
custom_settings.update({
"ideogram_mu": preset["mu"],
"ideogram_std": preset["std"],
})
_drop_legacy_cfg_override_settings(custom_settings)
ui_defaults.update({
"num_inference_steps": preset["num_inference_steps"],
"guidance_phases": 2,
"guidance_scale": 7.0,
"guidance2_scale": 3.0,
"switch_threshold": preset["switch_threshold"],
"sample_solver": _DEFAULT_SAMPLE_SOLVER,
"flow_shift": _DEFAULT_FLOW_SHIFT,
"custom_settings": custom_settings,
})
def _model_uses_nf4(model_def):
urls = []
if isinstance(model_def, dict):
urls.extend(model_def.get("URLs", []) or [])
urls.extend(model_def.get("URLs2", []) or [])
return any("_nf4" in str(url).lower() for url in urls)
def _conditional_transformer_only(base_model_type, model_def=None):
return base_model_type == _TURBOTIME_MODEL_TYPE or bool((model_def or {}).get("conditional_transformer_only", False))
class family_handler:
@staticmethod
def query_model_def(base_model_type, model_def):
conditional_only = _conditional_transformer_only(base_model_type, model_def)
text_encoder_filename = "Qwen3-VL-8B-Instruct_nf4.safetensors" if _model_uses_nf4(model_def) else "Qwen3-VL-8B-Instruct_fp8.safetensors"
model_def_update = {
"image_outputs": True,
"flux2": True,
"vae_upsamplers": {"flux2_vae_pid": [1]},
"excluded_spatial_upsamplers": ["flux2_pid"],
"guidance_max_phases": 3,
"lora_multiplier_branches": ["cond", "uncond"],
"inference_steps": True,
"fit_into_canvas_image_refs": 0,
"profiles_dir": [base_model_type],
"preset_profiles_dir": ["ideogram4_presets"],
"sample_solvers": _SAMPLE_SOLVERS,
"flow_shift": True,
"custom_settings": [one.copy() for one in _CUSTOM_SETTINGS],
"infos": IDEOGRAM4_INFOS,
"no_negative_prompt": True,
"no_background_removal": True,
"skip_prompt_template": True,
"resolutions_categories": ["<=2k"],
"vae_block_size": 16,
"prompt_enhancer_button_label": "Magic Prompt",
"prompt_infos": IDEOGRAM4_PROMPT_INFOS,
"prompt_helper_popup_dims": [78, 100],
"prompt_enhancer_def": {
"selection": ["T"],
"labels": {"T": "Ideogram JSON caption from Text Prompt"},
"default": "",
},
"text_prompt_enhancer_instructions": IDEOGRAM4_PROMPT_ENHANCER,
"image_prompt_enhancer_instructions": IDEOGRAM4_PROMPT_ENHANCER,
"text_prompt_enhancer_max_tokens": 2048,
"image_prompt_enhancer_max_tokens": 2048,
"text_encoder_folder": _TEXT_ENCODER_FOLDER,
"text_encoder_URLs": [
build_hf_url(_PROJECT_REPO, _TEXT_ENCODER_FOLDER, text_encoder_filename),
],
}
if conditional_only:
model_def_update.update({
"conditional_transformer_only": True,
"guidance_max_phases": 0,
"lora_multiplier_phases": 1,
"preset_profiles_dir": [],
"infos": IDEOGRAM4_TURBOTIME_INFOS,
})
model_def_update.pop("lora_multiplier_branches", None)
return model_def_update
@staticmethod
def query_supported_types():
return ["ideogram4", _TURBOTIME_MODEL_TYPE]
@staticmethod
def query_family_maps():
return {}, {}
@staticmethod
def query_model_family():
return "ideogram4"
@staticmethod
def query_family_infos():
return {"ideogram4": (1140, "Ideogram")}
@staticmethod
def render_prompt_helper(model_type, model_def, prompt_id, popup_id, prompt_elem_id, resolution_elem_id):
from .prompt_helper import render_prompt_helper
return render_prompt_helper(model_type, model_def, prompt_id, popup_id, prompt_elem_id, resolution_elem_id)
@staticmethod
def get_prompt_helper_css():
from .prompt_helper import get_prompt_helper_css
return get_prompt_helper_css()
@staticmethod
def get_prompt_helper_javascript():
from .prompt_helper import get_prompt_helper_javascript
return get_prompt_helper_javascript()
@staticmethod
def get_rgb_factors(base_model_type):
from shared.RGB_factors import get_rgb_factors
return get_rgb_factors("flux", sub_family="flux2")
@staticmethod
def register_lora_cli_args(parser, lora_root):
parser.add_argument(
"--lora-dir-ideogram4",
type=str,
default=None,
help=f"Path to a directory that contains Ideogram 4 LoRAs (default: {os.path.join(lora_root, 'ideogram4')})",
)
@staticmethod
def get_lora_dir(base_model_type, args, lora_root):
return getattr(args, "lora_dir_ideogram4", None) or os.path.join(lora_root, "ideogram4")
@staticmethod
def query_model_files(computeList, base_model_type, model_def=None):
transformer_configs = ["ideogram4_transformer_config.json"]
if not _conditional_transformer_only(base_model_type, model_def):
transformer_configs.append("ideogram4_unconditional_transformer_config.json")
return [
{
"repoId": _VAE_REPO,
"sourceFolderList": [""],
"fileList": [[_VAE_FILENAME]],
},
{
"repoId": _PROJECT_REPO,
"sourceFolderList": [_PROJECT_FOLDER, _TEXT_ENCODER_FOLDER],
"fileList": [
transformer_configs,
["config.json", "tokenizer.json", "tokenizer_config.json", "chat_template.jinja"],
],
}
]
@staticmethod
def load_model(
model_filename,
model_type=None,
base_model_type=None,
model_def=None,
quantizeTransformer=False,
text_encoder_quantization=None,
dtype=torch.bfloat16,
VAE_dtype=torch.float32,
mixed_precision_transformer=False,
save_quantized=False,
submodel_no_list=None,
text_encoder_filename=None,
**kwargs,
):
from .ideogram4_main import model_factory
pipe_processor = model_factory(
checkpoint_dir="ckpts",
model_filename=model_filename,
model_type=model_type,
model_def=model_def,
base_model_type=base_model_type,
text_encoder_filename=text_encoder_filename,
dtype=dtype,
VAE_dtype=VAE_dtype,
save_quantized=save_quantized,
)
pipe = {
"transformer": pipe_processor.conditional_transformer,
"text_encoder": pipe_processor.text_encoder,
"vae": pipe_processor.autoencoder,
}
co_tenants_map = {}
if pipe_processor.unconditional_transformer is not None:
pipe["transformer2"] = pipe_processor.unconditional_transformer
co_tenants_map = {
"transformer": ["transformer2"],
"transformer2": ["transformer"],
}
pipe = {
"pipe": pipe,
"coTenantsMap": co_tenants_map,
}
return pipe_processor, pipe
@staticmethod
def update_default_settings(base_model_type, model_def, ui_defaults):
ui_defaults.update({"image_mode": 1, "model_mode": None, "batch_size": 1})
if _conditional_transformer_only(base_model_type, model_def):
_apply_turbotime_settings(ui_defaults)
return
_apply_preset_settings(ui_defaults, _DEFAULT_PRESET)
@staticmethod
def fix_settings(base_model_type, settings_version, model_def, ui_defaults):
if _conditional_transformer_only(base_model_type, model_def):
_apply_turbotime_settings(ui_defaults, preserve_user_values=True)
return
old_mode = ui_defaults.get("model_mode", None)
if old_mode in _PRESET_SETTINGS:
_apply_preset_settings(ui_defaults, old_mode)
elif ui_defaults.get("sample_solver", None) in _PRESET_SETTINGS:
_apply_preset_settings(ui_defaults, ui_defaults["sample_solver"])
else:
custom_settings = ui_defaults.get("custom_settings", None)
if not isinstance(custom_settings, dict):
custom_settings = {}
defaults = _PRESET_SETTINGS[_DEFAULT_PRESET]
if custom_settings.get("ideogram_cfg_override_enabled", None) not in (None, 0, "0", False) and "ideogram_cfg_override_start_percent" in custom_settings:
ui_defaults["switch_threshold"] = _percent_to_switch_threshold(custom_settings["ideogram_cfg_override_start_percent"], ui_defaults.get("flow_shift", _DEFAULT_FLOW_SHIFT))
_drop_legacy_cfg_override_settings(custom_settings)
custom_settings.setdefault("ideogram_mu", defaults["mu"])
custom_settings.setdefault("ideogram_std", defaults["std"])
ui_defaults["custom_settings"] = custom_settings
ui_defaults.setdefault("num_inference_steps", defaults["num_inference_steps"])
ui_defaults.setdefault("guidance_scale", 7.0)
ui_defaults.setdefault("guidance2_scale", 3.0)
ui_defaults.setdefault("switch_threshold", defaults["switch_threshold"])
if ui_defaults.get("sample_solver", "") not in _SAMPLE_SOLVER_IDS:
ui_defaults["sample_solver"] = _DEFAULT_SAMPLE_SOLVER
ui_defaults.setdefault("flow_shift", _DEFAULT_FLOW_SHIFT)
ui_defaults.setdefault("guidance_phases", 2)
ui_defaults["model_mode"] = None
ui_defaults.setdefault("image_mode", 1)
|