Sync SeFi Image app from working Space

app.py

@@ -5,6 +5,7 @@ import gc
 import os
 import random
 import threading
+import time
 import traceback
 import warnings
 from dataclasses import dataclass
@@ -330,6 +331,14 @@ def _friendly_error(exc: BaseException, repo_id: str | None = None) -> str:
     return f"{type(exc).__name__}: {text}"
 
 
+def _format_seconds(seconds: float) -> str:
+    seconds = max(0, int(round(seconds)))
+    minutes, secs = divmod(seconds, 60)
+    if minutes:
+        return f"{minutes}m {secs:02d}s"
+    return f"{secs}s"
+
+
 def model_defaults(model_key: str):
     preset = MODEL_PRESETS[model_key]
     return (
@@ -369,6 +378,8 @@ def estimate_duration(
     guidance_scale: float,
     seed: int,
     randomize_seed: bool,
+    *_args,
+    **_kwargs,
 ) -> int:
     del prompt, guidance_scale, seed, randomize_seed
     preset = MODEL_PRESETS.get(model_key)
@@ -387,7 +398,9 @@ def generate(
     guidance_scale: float,
     seed: int,
     randomize_seed: bool,
+    progress=gr.Progress(track_tqdm=False),
 ):
+    request_started_at = time.monotonic()
     prompt = prompt.strip()
     if not prompt:
         return None, "Enter a prompt.", seed
@@ -415,7 +428,25 @@ def generate(
         if torch.cuda.is_available():
             torch.backends.cuda.matmul.allow_tf32 = True
 
+        progress(0, desc=f"Loading {preset.label}")
         pipe = _load_pipe(model_key)
+        denoise_started_at = time.monotonic()
+        progress(0, desc=f"Denoising 0/{steps} steps")
+
+        def report_step(step: int, total: int) -> None:
+            total = max(1, int(total))
+            step = min(max(0, int(step)), total)
+            elapsed = time.monotonic() - denoise_started_at
+            remaining = 0.0
+            if step > 0:
+                remaining = (elapsed / step) * (total - step)
+            desc = (
+                f"Denoising {step}/{total} steps | "
+                f"elapsed {_format_seconds(elapsed)} | "
+                f"ETA {_format_seconds(remaining)}"
+            )
+            progress(step / total, desc=desc)
+
         images = pipe(
             prompt,
             num_inference_steps=steps,
@@ -423,7 +454,9 @@ def generate(
             height=height,
             width=width,
             seed=int(seed),
+            progress_callback=report_step,
         )
+        progress(1, desc="Finalizing image")
     except Exception as exc:
         traceback.print_exc()
         return None, _friendly_error(exc, preset.repo_id), seed
@@ -435,7 +468,8 @@ def generate(
         images[0],
         (
             f"Generated with `{preset.repo_id}` at {width}x{height}, "
-            f"{steps} steps, guidance {guidance_scale}, seed {seed}."
+            f"{steps} steps, guidance {guidance_scale}, seed {seed}. "
+            f"Total time: {_format_seconds(time.monotonic() - request_started_at)}."
         ),
         seed,
     )

requirements.txt

@@ -1,7 +1,7 @@
 gradio==6.19.0
 spaces>=0.50.0
-torch>=2.9.1
-torchvision>=0.24.0
+torch==2.9.1
+torchvision==0.24.1
 diffusers>=0.39.0
 transformers>=5.13.0
 accelerate>=1.12.0

sefi/checkpoints.py

@@ -9,6 +9,10 @@ from pathlib import Path
 CONFIG_FILENAMES = ("sefi_config.yaml", "config.yaml")
 
 
+def _hub_token() -> str | None:
+    return os.getenv("HF_TOKEN") or os.getenv("HUGGING_FACE_HUB_TOKEN")
+
+
 def _download_hf_snapshot(
     repo_id: str,
     *,
@@ -26,6 +30,7 @@ def _download_hf_snapshot(
         repo_id=repo_id,
         cache_dir=str(cache_dir),
         local_files_only=False,
+        token=_hub_token(),
     )
 
 

sefi/pipeline.py

@@ -3,7 +3,7 @@
 from __future__ import annotations
 
 from pathlib import Path
-from typing import Iterable
+from typing import Callable, Iterable
 
 import torch
 from PIL import Image
@@ -106,6 +106,7 @@ class SEFIInferencePipeline:
         batch_size: int | None = None,
         seed: int | None = None,
         generator: torch.Generator | None = None,
+        progress_callback: Callable[[int, int], None] | None = None,
     ) -> list[Image.Image]:
         prompt_list = [prompts] if isinstance(prompts, str) else list(prompts)
         if not prompt_list:
@@ -156,6 +157,7 @@ class SEFIInferencePipeline:
                     height=size.height,
                     width=size.width,
                     generator=gen,
+                    progress_callback=progress_callback,
                 )
             )
         return images

sefi/runner.py

@@ -5,7 +5,7 @@ from __future__ import annotations
 import json
 import math
 import os
-from typing import Optional
+from typing import Callable, Optional
 
 import torch
 from PIL import Image
@@ -606,6 +606,7 @@ class SEFIInferenceRunner:
         height: int,
         width: int,
         generator: Optional[torch.Generator] = None,
+        progress_callback: Optional[Callable[[int, int], None]] = None,
     ) -> list[Image.Image]:
         if num_inference_steps <= 0:
             raise ValueError("num_inference_steps must be > 0")
@@ -784,6 +785,8 @@ class SEFIInferenceRunner:
             lat_sem = lat_sem + dt_sem * vel_sem
             lat_tex = lat_tex + dt_tex * vel_tex
             latents = torch.cat([lat_sem, lat_tex], dim=1)
+            if progress_callback is not None:
+                progress_callback(step + 1, num_inference_steps)
 
         texture_latents = latents[:, self.semantic_channels :]
         decoded = self.texture_codec.decode_texture(