adam/tools/flow_generator.py

"""Image-generation bridge for the connected Flow Matching project."""

from __future__ import annotations

import importlib.util
import json
import random
import re
import sys
from datetime import datetime, timezone
from pathlib import Path
from types import ModuleType

from adam.config import ConfigManager
from adam.executor import ToolCancelled, ToolContext, ToolExecutionError
from adam.generations import generation_metadata_path, generation_output_folder
from adam.generation_previews import accepts_preview_callback, publish_generation_preview
from adam.image_preferences import GenerationPreferenceEvaluator, PreferenceProfile


_backend_module: ModuleType | None = None
_backend_script: Path | None = None
_loaded_model: object | None = None
_loaded_model_path: Path | None = None


def _load_backend(script: Path) -> ModuleType:
    global _backend_module, _backend_script
    if _backend_module is not None and _backend_script == script:
        return _backend_module
    module_name = "_adam_connected_flow_generator"
    spec = importlib.util.spec_from_file_location(module_name, script)
    if spec is None or spec.loader is None:
        raise ToolExecutionError("The connected Flow Matching generator could not be loaded.")
    module = importlib.util.module_from_spec(spec)
    sys.modules[module_name] = module
    try:
        spec.loader.exec_module(module)
    except Exception as exc:
        sys.modules.pop(module_name, None)
        raise ToolExecutionError(f"Could not load the Flow Matching generator: {exc}") from exc
    if not callable(getattr(module, "load_unet", None)) or not callable(
        getattr(module, "sample_flow", None)
    ):
        raise ToolExecutionError(
            "The connected Flow Matching app does not expose its generation functions."
        )
    _backend_module = module
    _backend_script = script
    return module


def _safe_label(value: str) -> str:
    label = value.strip()[:96] or "Flow"
    label = re.sub(r"[<>:\"/\\|?*\x00-\x1f]+", " ", label)
    label = re.sub(r"\s+", " ", label).strip(" .")
    return label or "Flow"


def generate_flow_images(
    context: ToolContext,
    model_name: str,
    model_path: str,
    prompt: str,
    image_count: int,
    steps: int,
    seed: int,
    sampler: str,
    aspect_ratio: str,
    width: int = 0,
    height: int = 0,
    preview_interval: int = 0,
    smart_generation: bool = False,
    smart_wanted_results: int = 0,
    smart_max_candidates: int = 0,
    smart_min_score: float = 0.70,
    smart_mode: str = "threshold",
    smart_keep_rejected: bool = True,
) -> dict[str, object]:
    global _loaded_model, _loaded_model_path
    config = ConfigManager(context.root)
    flow_root = Path(
        str(config.get("tool_folders", {}).get("flow_trainer", ""))
    ).expanduser().resolve()
    script = flow_root / "flow_matching_app.py"
    if not script.is_file():
        raise ToolExecutionError(
            "Flow Matching flow_matching_app.py was not found. Re-scan its folder in Settings."
        )
    model = Path(model_path).expanduser().resolve()
    allowed_root = (flow_root / "output_flow_models").resolve()
    try:
        model.relative_to(allowed_root)
    except ValueError as exc:
        raise ToolExecutionError(
            "The Flow model must be inside the connected Flow Matching output folder."
        ) from exc
    try:
        info = json.loads((model / "flow_model_info.json").read_text(encoding="utf-8"))
    except (OSError, ValueError, TypeError, json.JSONDecodeError) as exc:
        raise ToolExecutionError("Choose a completed Flow Matching image model.") from exc
    if info.get("model_type") != "rectified_flow" or not (
        model / "unet" / "config.json"
    ).is_file():
        raise ToolExecutionError("Choose a completed Flow Matching image model.")

    missing_packages = [
        package
        for package in ("torch", "torchvision", "diffusers", "PIL")
        if importlib.util.find_spec(package) is None
    ]
    if missing_packages:
        raise ToolExecutionError(
            "ADAM's Python environment is missing Flow generation packages: "
            + ", ".join(missing_packages)
            + ". Install the connected Flow Matching requirements, then restart ADAM."
        )
    count = int(image_count)
    step_count = int(steps)
    smart_enabled = bool(smart_generation)
    wanted_results = int(smart_wanted_results or count)
    max_candidates = int(smart_max_candidates or count)
    if smart_enabled:
        if not 1 <= wanted_results <= 48:
            raise ToolExecutionError("Wanted Smart Generation results must be between 1 and 48.")
        if not wanted_results <= max_candidates <= 256:
            raise ToolExecutionError("Maximum Smart Generation candidates must be between wanted results and 256.")
        if not 0.0 <= float(smart_min_score) <= 1.0:
            raise ToolExecutionError("Minimum Smart Generation score must be between 0.00 and 1.00.")
        if str(smart_mode).casefold() not in {"threshold", "top_n"}:
            raise ToolExecutionError("Smart Generation mode must be threshold or top_n.")
        count = wanted_results
    if not 1 <= count <= 48:
        raise ToolExecutionError("Image count must be between 1 and 48.")
    if not 1 <= step_count <= 200:
        raise ToolExecutionError("Flow steps must be between 1 and 200.")
    method = sampler.strip().title()
    if method not in {"Heun", "Euler"}:
        raise ToolExecutionError("Flow generation supports the Heun and Euler methods.")
    allowed_aspects = {
        "1:1 (Square)", "4:3 (Landscape)", "3:4 (Portrait)",
        "3:2 (Landscape)", "2:3 (Portrait)", "16:9 (Widescreen)",
        "9:16 (Vertical)",
    }
    if aspect_ratio not in allowed_aspects:
        raise ToolExecutionError("Choose one of the supported Flow aspect ratios.")
    custom_width = int(width or 0)
    custom_height = int(height or 0)
    if bool(custom_width) != bool(custom_height):
        raise ToolExecutionError("Set both Flow image width and height, or leave both unset.")
    if custom_width and (
        not 64 <= custom_width <= 2048
        or not 64 <= custom_height <= 2048
        or custom_width % 16
        or custom_height % 16
    ):
        raise ToolExecutionError(
            "Flow image width and height must each be between 64 and 2048 pixels and divisible by 16."
        )
    if len(prompt) > 500:
        raise ToolExecutionError("The generation label must be 500 characters or shorter.")
    if not 0 <= int(preview_interval) <= step_count:
        raise ToolExecutionError("Preview interval must be between 0 and the total number of steps.")
    base_seed = int(seed)
    if base_seed <= 0:
        base_seed = random.randint(1, 2_147_483_647 - count)
    seed_count = max_candidates if smart_enabled else count
    if base_seed + seed_count - 1 > 2_147_483_647:
        raise ToolExecutionError("The seed is too large for this image count.")

    backend = _load_backend(script)
    preview_enabled = int(preview_interval) > 0
    preview_supported = accepts_preview_callback(backend.sample_flow)
    if preview_enabled and not preview_supported:
        context.log("This connected Flow generator does not yet expose denoising previews; generation will continue normally.")
    try:
        import torch

        device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
        dtype = torch.float16 if device.type == "cuda" else torch.float32
        if _loaded_model is None or _loaded_model_path != model:
            _loaded_model = None
            if torch.cuda.is_available():
                torch.cuda.empty_cache()
            context.log(f"Loading completed Flow Matching model: {model_name}")
            _loaded_model = backend.load_unet(model, device=device, dtype=dtype)
            _loaded_model_path = model
    except ToolCancelled:
        raise
    except Exception as exc:
        raise ToolExecutionError(f"Could not load the Flow Matching model: {exc}") from exc

    timestamp = datetime.now(timezone.utc).strftime("%Y%m%d_%H%M%S")
    output = generation_output_folder(context.root, context.tool.id, _safe_label(model_name))
    context.log("Flow Matching models generate learned visual samples; the label is metadata, not a text prompt.")
    image_paths: list[str] = []
    image_evaluations: dict[str, dict[str, object]] = {}
    selected_paths: list[str] = []
    generated_total = max_candidates if smart_enabled else count
    profile = PreferenceProfile(context.root, context.tool.id, _safe_label(model_name), str(model)) if smart_enabled else None
    evaluator = GenerationPreferenceEvaluator(context.root) if smart_enabled else None
    threshold = float(smart_min_score)
    top_n_mode = str(smart_mode).casefold() == "top_n"
    for index in range(generated_total):
        context.checkpoint()
        current_seed = base_seed + index

        def on_progress(done: int, total: int, image_index: int = index) -> None:
            completed = image_index + (done / max(1, total))
            context.progress(
                max(1, min(99, round(completed * 100 / generated_total))),
                f"Generating image {image_index + 1} of {generated_total} · flow step {done} of {total}" if smart_enabled else f"Generating image {image_index + 1} of {count} · flow step {done} of {total}",
            )

        try:
            settings = {"aspect_ratio": aspect_ratio}
            if custom_width:
                settings.update({"width": custom_width, "height": custom_height})
            if preview_enabled and preview_supported:
                settings["preview_interval"] = int(preview_interval)
                settings["preview_callback"] = lambda payload, step=0, total_steps=step_count, current=index: publish_generation_preview(
                    context, output, payload, image_index=current, image_count=count,
                    step=step, total_steps=total_steps,
                )
            images = backend.sample_flow(
                _loaded_model,
                1,
                step_count,
                device,
                dtype,
                current_seed,
                method,
                on_progress,
                **settings,
            )
            destination = output / f"{timestamp}_{context.job_id}_{method}_seed_{current_seed}.png"
            images[0].save(destination, format="PNG")
        except ToolCancelled:
            raise
        except Exception as exc:
            raise ToolExecutionError(f"Flow Matching generation failed: {exc}") from exc
        image_paths.append(str(destination))
        if smart_enabled and profile is not None and evaluator is not None:
            score = evaluator.score(profile, [destination], keep_threshold=threshold, reject_threshold=profile.reject_threshold)[0]
            image_evaluations[str(destination.resolve())] = {
                "score": score.score,
                "confidence": score.confidence,
                "category": score.category,
                "reason": score.reason,
            }
            if not top_n_mode and score.score is not None and score.score >= threshold:
                selected_paths.append(str(destination))
                if len(selected_paths) >= wanted_results:
                    break

    created_at = datetime.now(timezone.utc).isoformat()
    if smart_enabled and top_n_mode:
        ranked = sorted(
            image_paths,
            key=lambda path: float(image_evaluations.get(str(Path(path).resolve()), {}).get("score") or -1.0),
            reverse=True,
        )
        selected_paths = ranked[:wanted_results]
    if smart_enabled:
        ordered_images = [*selected_paths, *[path for path in image_paths if path not in set(selected_paths)]]
    else:
        ordered_images = image_paths
    metadata = {
        "version": 1,
        "provider_id": context.tool.id,
        "provider_name": context.tool.name,
        "model_name": _safe_label(model_name),
        "model_path": str(model),
        "prompt": prompt.strip(),
        "prompt_behavior": "label_only",
        "seed": base_seed,
        "image_seeds": [base_seed + index for index in range(count)],
        "image_count": len(ordered_images) if smart_enabled and smart_keep_rejected else count,
        "steps": step_count,
        "sampler": method,
        "aspect_ratio": aspect_ratio,
        "width": custom_width,
        "height": custom_height,
        "preview_interval": int(preview_interval),
        "preview_supported": preview_supported,
        "images": ordered_images if smart_keep_rejected or not smart_enabled else selected_paths,
        "image_evaluations": image_evaluations,
        "smart_generation": {
            "enabled": smart_enabled,
            "mode": str(smart_mode),
            "wanted_results": wanted_results if smart_enabled else count,
            "maximum_candidates": max_candidates if smart_enabled else count,
            "minimum_score": threshold,
            "selected_count": len(selected_paths) if smart_enabled else count,
            "candidate_count": len(image_paths),
            "profile_id": profile.id if profile else "",
            "keep_rejected_candidates": bool(smart_keep_rejected),
        },
        "created_at": created_at,
    }
    generation_metadata_path(output, timestamp, context.job_id).write_text(
        json.dumps(metadata, indent=2), encoding="utf-8"
    )
    if evaluator is not None:
        evaluator.vision.unload()
    if smart_enabled:
        context.progress(100, f"Smart Generation selected {len(selected_paths)} of {wanted_results} requested image(s) from {len(image_paths)} candidate(s)")
    else:
        context.progress(100, f"Generated {count} image(s)")
    return {
        "output_folder": str(output),
        "assets": [
            {
                "kind": "generation",
                "name": f"{_safe_label(model_name)} · {timestamp}",
                "path": str(output),
                "trainer": "flow",
            }
        ],
    }