File size: 7,351 Bytes
32c5da4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
from __future__ import annotations

import logging
import os
from pathlib import Path

from .interface import ProviderRequest, ProviderResult, ProviderUnavailableError

LOGGER = logging.getLogger(__name__)

# Lazy imports to avoid torch loading issues on Windows
torch = None
StableDiffusionImg2ImgPipeline = None
StableDiffusionPipeline = None

def _ensure_imports():
    global torch, StableDiffusionImg2ImgPipeline, StableDiffusionPipeline
    if torch is not None:
        return
    try:
        import torch as _torch
        from diffusers import StableDiffusionImg2ImgPipeline as _Img2Img
        from diffusers import StableDiffusionPipeline as _Pipeline
        torch = _torch
        StableDiffusionImg2ImgPipeline = _Img2Img
        StableDiffusionPipeline = _Pipeline
    except Exception:  # pragma: no cover - optional dependency
        pass


class DiffusionProvider:
    id = "diffusion"
    name = "Stable Diffusion (local)"
    description = "Uses diffusers for local Stable Diffusion generation"

    def __init__(self, model_id: str = "segmind/tiny-sd") -> None:
        self.model_id = os.getenv("IMAGEFORGE_DIFFUSION_MODEL", model_id)
        self._pipe: StableDiffusionPipeline | None = None
        self._img2img_pipe: StableDiffusionImg2ImgPipeline | None = None

    def is_available(self) -> bool:
        _ensure_imports()
        return StableDiffusionPipeline is not None and torch is not None

    def _ensure_pipeline(self) -> StableDiffusionPipeline:
        _ensure_imports()
        if StableDiffusionPipeline is None or torch is None:
            raise ProviderUnavailableError(
                "Diffusion dependencies missing. Install diffusers, torch, and transformers."
            )
        if self._pipe is None:
            device = "cuda" if torch.cuda.is_available() else "cpu"
            dtype = torch.float16 if device == "cuda" else torch.float32
            local_only = os.getenv("IMAGEFORGE_DIFFUSION_LOCAL_ONLY", "0") == "1"
            LOGGER.info("Loading diffusion model '%s' on %s", self.model_id, device)
            try:
                pipe = StableDiffusionPipeline.from_pretrained(
                    self.model_id,
                    torch_dtype=dtype,
                    local_files_only=local_only,
                )
            except Exception as exc:  # noqa: BLE001
                mode_hint = "local cache only" if local_only else "online download"
                raise ProviderUnavailableError(
                    f"Diffusion model '{self.model_id}' could not be loaded ({mode_hint}). "
                    "Set IMAGEFORGE_DIFFUSION_LOCAL_ONLY=0 to allow downloading models."
                ) from exc
            if device == "cuda":
                pipe = pipe.to(device)
                if os.getenv("IMAGEFORGE_ENABLE_ATTENTION_SLICING", "1") == "1":
                    pipe.enable_attention_slicing()
            self._pipe = pipe
        return self._pipe

    def _ensure_img2img_pipeline(self) -> StableDiffusionImg2ImgPipeline | None:
        _ensure_imports()
        if StableDiffusionImg2ImgPipeline is None or torch is None:
            return None
        if self._img2img_pipe is None:
            device = "cuda" if torch.cuda.is_available() else "cpu"
            dtype = torch.float16 if device == "cuda" else torch.float32
            local_only = os.getenv("IMAGEFORGE_DIFFUSION_LOCAL_ONLY", "0") == "1"
            LOGGER.info("Loading diffusion img2img model '%s' on %s", self.model_id, device)
            try:
                img2img = StableDiffusionImg2ImgPipeline.from_pretrained(
                    self.model_id,
                    torch_dtype=dtype,
                    local_files_only=local_only,
                )
            except Exception as exc:  # noqa: BLE001
                mode_hint = "local cache only" if local_only else "online download"
                raise ProviderUnavailableError(
                    f"Diffusion img2img model '{self.model_id}' could not be loaded ({mode_hint}). "
                    "Set IMAGEFORGE_DIFFUSION_LOCAL_ONLY=0 to allow downloading models."
                ) from exc
            if device == "cuda":
                img2img = img2img.to(device)
                if os.getenv("IMAGEFORGE_ENABLE_ATTENTION_SLICING", "1") == "1":
                    img2img.enable_attention_slicing()
            self._img2img_pipe = img2img
        return self._img2img_pipe

    def generate(self, request: ProviderRequest, output_dir: Path, progress, is_cancelled) -> ProviderResult:
        if is_cancelled():
            return ProviderResult(image_paths=[])
        if request.model_variant and request.model_variant != self.model_id:
            self.model_id = request.model_variant
            self._pipe = None
            self._img2img_pipe = None
        progress(1, "Loading diffusion model")
        pipe = self._ensure_pipeline()
        output_dir.mkdir(parents=True, exist_ok=True)
        image_paths: list[Path] = []

        for idx in range(request.count):
            if is_cancelled():
                break

            seed = request.seed + idx
            generator = torch.Generator(device=pipe.device).manual_seed(seed)

            def _callback(step: int, timestep: int, latents):  # noqa: ANN001
                if is_cancelled():
                    raise RuntimeError("Generation cancelled")
                local_progress = int(((step + 1) / max(1, request.steps)) * 100)
                progress(local_progress, f"Diffusion step {step + 1}/{request.steps} (image {idx + 1})")

            if request.init_image_path:
                from PIL import Image

                img2img_pipe = self._ensure_img2img_pipeline()
                if img2img_pipe is None:
                    raise ProviderUnavailableError("Img2Img requires diffusers img2img pipeline support")
                init_image = Image.open(request.init_image_path).convert("RGB").resize((request.width, request.height))
                result = img2img_pipe(
                    prompt=request.prompt,
                    negative_prompt=request.negative_prompt or None,
                    image=init_image,
                    num_inference_steps=request.steps,
                    guidance_scale=request.guidance,
                    strength=max(0.0, min(1.0, request.img2img_strength)),
                    generator=generator,
                    callback=_callback,
                    callback_steps=1,
                )
            else:
                result = pipe(
                    prompt=request.prompt,
                    negative_prompt=request.negative_prompt or None,
                    width=request.width,
                    height=request.height,
                    num_inference_steps=request.steps,
                    guidance_scale=request.guidance,
                    generator=generator,
                    callback=_callback,
                    callback_steps=1,
                )
            image = result.images[0]
            image_path = output_dir / f"image_{idx + 1:02d}.png"
            image.save(image_path, format="PNG")
            image_paths.append(image_path)
            pct = int(((idx + 1) / request.count) * 100)
            progress(pct, f"Diffusion image {idx + 1}/{request.count} complete")

        return ProviderResult(image_paths=image_paths)