Use generation 2 names and seed

API.md

@@ -19,7 +19,7 @@ Content-Type: application/json
 {"positive":"an orange cat","negative":"text, watermark"}
 ```
 
-Queues a passive confusion-matrix run and returns its reserved `grN.epng` path. Each numbered checkpoint gets one 20-step first pass at CFG 6 with DPM++ 2M and Karras. Diagonal `gN.epng` files contain that first pass. Each missing off-diagonal `gN-M.epng` resumes from the first-pass image, applies a 1.1 bislerp latent upscale, then runs the second checkpoint for 14 steps at CFG 5 and denoise 0.35 with the same sampler, scheduler, prompts, and aDetailer. Existing cell files are reused. The labeled grid is saved after every cell is present.
+Queues a passive confusion-matrix run and returns its reserved `Ngr2.epng` path. Each numbered checkpoint gets one 20-step first pass at CFG 6 with DPM++ 2M and Karras. Diagonal `Ng2.epng` files contain that first pass. Each missing off-diagonal `N-Mg2.epng` resumes from the first-pass image, applies a 1.1 bislerp latent upscale, then runs the second checkpoint for 14 steps at CFG 5 and denoise 0.35 with the same sampler, scheduler, prompts, and aDetailer. Existing cell files are reused. The labeled grid is saved after every cell is present.
 
 ## Generate
 
@@ -52,7 +52,7 @@ Content-Type: application/json
 }
 ```
 
-Returns a metadata-free PNG scaled after generation by the decimal `return_scale` from `0.01` through `1`, with content type `image/png`. Use `1` for full resolution. Set either scheduler to `AlignYourSteps` to use `AlignYourStepsScheduler` with fixed model type `SDXL` and the corresponding steps and denoise values. Only `prompt` is required for the built-in site defaults. LoRAs run in array order. Each request creates a random first-pass seed and, when `upscale` is true, performs latent upscaling followed by a second sampler with another random seed. Leave `second_model` empty to reuse the first model and LoRA chain; otherwise `second_loras` applies to the selected second model. Width and height must be multiples of 8 from 64 through 2048; steps must be from 1 through 100.
+Returns a metadata-free PNG scaled after generation by the decimal `return_scale` from `0.01` through `1`, with content type `image/png`. Use `1` for full resolution. Set either scheduler to `AlignYourSteps` to use `AlignYourStepsScheduler` with fixed model type `SDXL` and the corresponding steps and denoise values. Only `prompt` is required for the built-in site defaults. LoRAs run in array order. Every sampler uses generation seed `2`. When `upscale` is true, the request performs latent upscaling followed by a second sampler. Leave `second_model` empty to reuse the first model and LoRA chain; otherwise `second_loras` applies to the selected second model. Width and height must be multiples of 8 from 64 through 2048; steps must be from 1 through 100.
 
 ## Archive
 

app.py

@@ -1,5 +1,4 @@
 import os
-import random
 import re
 import secrets
 import shutil
@@ -50,7 +49,12 @@ from PIL import Image, ImageDraw, ImageFilter, ImageFont
 from PIL.PngImagePlugin import PngInfo
 from pydantic import BaseModel, ConfigDict, Field
 from starlette.background import BackgroundTask
-from workflow_api import ALIGN_MODEL_TYPE, ALIGN_SCHEDULER, image_metadata
+from workflow_api import (
+    ALIGN_MODEL_TYPE,
+    ALIGN_SCHEDULER,
+    GENERATION,
+    image_metadata,
+)
 
 
 DATA_MOUNT = Path("/data")
@@ -473,7 +477,7 @@ def generate_core(body):
             height=request.height,
             batch_size=1,
         )[0]
-        first_seed = random.randrange(2**64)
+        first_seed = GENERATION
         samples = run_sampler(
             model,
             first_seed,
@@ -502,7 +506,7 @@ def generate_core(body):
                 positive = state["encode"].encode(clip=clip, text=request.prompt)[0]
                 negative = state["encode"].encode(clip=clip, text=request.negative)[0]
 
-            second_seed = random.randrange(2**64)
+            second_seed = GENERATION
             samples = run_sampler(
                 model,
                 second_seed,
@@ -698,7 +702,7 @@ def run_adetailer(image, model, clip, negative, sampler, scheduler):
             pixels=image,
         )[0]
         latent["noise_mask"] = mask
-        seed = random.randrange(2**64)
+        seed = GENERATION
         samples = run_sampler(
             model,
             seed,
@@ -752,7 +756,7 @@ def generate_matrix_cell_core(body):
         folder = (IMAGE_DIR / request.folder).resolve()
         if folder.parent != IMAGE_DIR.resolve():
             raise ValueError("Invalid matrix folder")
-        diagonal = folder / f"g{first_id}.epng"
+        diagonal = folder / f"{first_id}g{GENERATION}.epng"
         state["vae"] = load_vae(DEFAULT_VAE)
 
         if first_id == second_id:
@@ -774,7 +778,7 @@ def generate_matrix_cell_core(body):
             )[0]
             samples = run_sampler(
                 model,
-                random.randrange(2**64),
+                GENERATION,
                 MATRIX_FIRST_STEPS,
                 MATRIX_FIRST_CFG,
                 MATRIX_SAMPLER,
@@ -792,7 +796,7 @@ def generate_matrix_cell_core(body):
             release_model(first_name)
             return first_id
 
-        output = folder / f"g{first_id}-{second_id}.epng"
+        output = folder / f"{first_id}-{second_id}g{GENERATION}.epng"
         if output.is_file():
             return f"{first_id}-{second_id}"
 
@@ -816,7 +820,7 @@ def generate_matrix_cell_core(body):
         )[0]
         samples = run_sampler(
             model,
-            random.randrange(2**64),
+            GENERATION,
             MATRIX_SECOND_STEPS,
             MATRIX_SECOND_CFG,
             MATRIX_SAMPLER,
@@ -1035,7 +1039,7 @@ def reserve_matrix_grid(folder):
     with matrix_lock:
         number = 1
         while (
-            (path / f"gr{number}.epng").exists()
+            (path / f"{number}gr{GENERATION}.epng").exists()
             or (folder, number) in matrix_grids
         ):
             number += 1
@@ -1065,9 +1069,9 @@ def matrix_models():
 
 def matrix_image_path(folder, first_id, second_id):
     name = (
-        f"g{first_id}.epng"
+        f"{first_id}g{GENERATION}.epng"
         if first_id == second_id
-        else f"g{first_id}-{second_id}.epng"
+        else f"{first_id}-{second_id}g{GENERATION}.epng"
     )
     return IMAGE_DIR / folder / name
 
@@ -1115,7 +1119,7 @@ def create_matrix_grid(folder, number, models):
                 ),
             )
 
-    output = IMAGE_DIR / folder / f"gr{number}.epng"
+    output = IMAGE_DIR / folder / f"{number}gr{GENERATION}.epng"
     save_named_image(grid, output)
 
 
@@ -1399,7 +1403,7 @@ def start_matrix(body: MatrixRequest):
     matrix_pool.submit(run_matrix, body, folder, number)
     return {
         "status": "started",
-        "grid": f"{folder}/gr{number}.epng",
+        "grid": f"{folder}/{number}gr{GENERATION}.epng",
     }
 
 

workflow_api.py

@@ -1,12 +1,12 @@
 # Imports
 import json
 import os
-import random
 import sys
 from typing import Sequence, Mapping, Any, Union
 
 ALIGN_MODEL_TYPE = "SDXL"
 ALIGN_SCHEDULER = "AlignYourSteps"
+GENERATION = 2
 
 
 def get_value_at_index(obj: Union[Sequence, Mapping], index: int) -> Any:
@@ -802,7 +802,7 @@ def main(unload_models: bool | None = None):
             ksampler = KSampler()
             vaedecode = VAEDecode()
             for q in range(1):
-                node_5_seed = prompt["5"]["inputs"]["seed"] = random.randint(1, 2**64)
+                node_5_seed = prompt["5"]["inputs"]["seed"] = GENERATION
                 ksampler_5 = ksampler.sample(
                     seed=node_5_seed,
                     steps=16,