Generate configurable image batches

API.md

@@ -50,6 +50,7 @@ Content-Type: application/json
   "negative": "text, watermark",
   "width": 1152,
   "height": 896,
+  "batch_size": 2,
   "model": "52_novaAnimeXL_ilV190.safetensors",
   "loras": [{"name": "8_bikabaka.safetensors", "strength": 1, "clip": 0}],
   "sampler": "euler_ancestral",
@@ -70,7 +71,7 @@ Content-Type: application/json
 }
 ```
 
-Returns an encrypted `.epng` with ComfyUI workflow and parameter metadata, scaled after generation by the decimal `return_scale` from `0.01` through `1`, with content type `application/octet-stream`. Regional prompts and their mask nodes are included in the metadata. Use `1` for full resolution. `regions` accepts up to three spatial prompts on a 5 by 5 grid. Set `regional_mode` to `attention` for Attention Couple (PPM), or omit it for soft conditioning. Regional prompts establish subjects from the first sampling step. The reduced-strength global prompt starts after the first fifth to complete the environment while regions remain active. Each internal region edge feathers by half a grid cell, so overlap adjacent regions by one cell for a smooth crossfade. `area` accepts `auto`, `full`, `tl`, `tc`, `tr`, `ml`, `mc`, `mr`, `bl`, `bc`, `br`, `th`, `mh`, `bh`, `lh`, `ch`, or `rh`. It also accepts any single cell from `a1` through `e5`, or a rectangular inclusive range such as `a1:c5`. `auto` places one prompt in the center, two left and right, or three left, center, and right. Columns run left to right and rows run top to bottom. Region strength defaults to `1`. Use the global `prompt` for environment, lighting, shared style, and composition outside regions. Omitting `regions` keeps whole-image generation unchanged. 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 and detailers run in array order. A blank detailer `prompt` or `negative` inherits the matching main prompt. Each request creates fresh sampler seeds. When `upscale` is true, the selected upscaler is normalized to the requested `upscale_scale` resolution before the 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.
+Returns an encrypted `.epng` with ComfyUI workflow and parameter metadata, scaled after generation by the decimal `return_scale` from `0.01` through `1`, with content type `application/octet-stream`. `batch_size` generates from 1 through 8 images at once and defaults to `2`. Multiple landscape images are stacked top to bottom; portrait or square images are joined left to right. Regional prompts and their mask nodes are included in the metadata. Use `1` for full resolution. `regions` accepts up to three spatial prompts on a 5 by 5 grid. Set `regional_mode` to `attention` for Attention Couple (PPM), or omit it for soft conditioning. Regional prompts establish subjects from the first sampling step. The reduced-strength global prompt starts after the first fifth to complete the environment while regions remain active. Each internal region edge feathers by half a grid cell, so overlap adjacent regions by one cell for a smooth crossfade. `area` accepts `auto`, `full`, `tl`, `tc`, `tr`, `ml`, `mc`, `mr`, `bl`, `bc`, `br`, `th`, `mh`, `bh`, `lh`, `ch`, or `rh`. It also accepts any single cell from `a1` through `e5`, or a rectangular inclusive range such as `a1:c5`. `auto` places one prompt in the center, two left and right, or three left, center, and right. Columns run left to right and rows run top to bottom. Region strength defaults to `1`. Use the global `prompt` for environment, lighting, shared style, and composition outside regions. Omitting `regions` keeps whole-image generation unchanged. 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 and detailers run in array order. A blank detailer `prompt` or `negative` inherits the matching main prompt. Each request creates fresh sampler seeds. When `upscale` is true, the selected upscaler is normalized to the requested `upscale_scale` resolution before the 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

@@ -101,6 +101,8 @@ IMAGE_SUFFIXES = (".epng",)
 PREVIEW_QUALITY = 75
 PREVIEW_SCALE = .25
 DEFAULT_RETURN_SCALE = 1
+DEFAULT_BATCH_SIZE = 2
+MAX_BATCH_SIZE = 8
 ENVIRONMENT_START = .2
 REGIONAL_GLOBAL_STRENGTH = .6
 CUSTOM_NODE_REPOS = {
@@ -211,7 +213,7 @@ MATRIX_TYPES = (
     "sampler+scheduler",
     COMBINED_MATRIX_TYPE,
 )
-GPU_DURATION = 13
+GPU_DURATION = 18
 GPU_ATTEMPTS = 3
 GPU_RETRY_ERRORS = (
     "gpu task aborted",
@@ -416,6 +418,7 @@ class DirectRequest(BaseModel):
     negative: str = DEFAULT_NEGATIVE
     width: int = Field(1152, ge=64, le=2048)
     height: int = Field(896, ge=64, le=2048)
+    batch_size: int = Field(DEFAULT_BATCH_SIZE, ge=1, le=MAX_BATCH_SIZE)
     sampler: str = DEFAULT_SAMPLER
     scheduler: str = DEFAULT_SCHEDULER
     steps: int = Field(DEFAULT_STEPS, ge=1, le=100)
@@ -1031,14 +1034,14 @@ def infer_image(
                 image=image,
             )[0]
 
-        return image[0]
+        return image
 
 
 def generate_gpu(request, regions):
     latent = state["latent"].generate(
         width=request.width,
         height=request.height,
-        batch_size=1,
+        batch_size=request.batch_size,
     )[0]
     first_seed = secrets.randbits(64)
     second_seed = secrets.randbits(64) if request.upscale else None
@@ -1492,7 +1495,7 @@ def model_options(kind, local):
     return sorted(set(local) | remote_models[kind].keys(), key=str.casefold)
 
 
-def generate_image(request, from_api=True):
+def generate_images(request, from_api=True):
     if (
         request.width % 8
         or request.height % 8
@@ -1533,34 +1536,60 @@ def generate_image(request, from_api=True):
             regions,
         )
 
-        result = Image.fromarray(
-            np.clip(
-                image.detach().numpy() * 255,
-                0,
-                255,
-            ).astype(np.uint8)
+        metadata = image_metadata(
+            request.model_dump(),
+            [first_seed, second_seed],
+            detailer_seeds,
+            [
+                vae_name(detailer.model)
+                if detailer.model
+                else second_vae if request.upscale else first_vae
+                for detailer in request.detailers
+            ],
+            [first_vae, second_vae],
+            regions,
+            ENVIRONMENT_START,
+            REGIONAL_GLOBAL_STRENGTH,
         )
-
-        result.info.update(
-            image_metadata(
-                request.model_dump(),
-                [first_seed, second_seed],
-                detailer_seeds,
-                [
-                    vae_name(detailer.model)
-                    if detailer.model
-                    else second_vae if request.upscale else first_vae
-                    for detailer in request.detailers
-                ],
-                [first_vae, second_vae],
-                regions,
-                ENVIRONMENT_START,
-                REGIONAL_GLOBAL_STRENGTH,
+        results = [
+            Image.fromarray(
+                np.clip(item.detach().numpy() * 255, 0, 255).astype(np.uint8)
             )
+            for item in image
+        ]
+        for result in results:
+            result.info.update(metadata)
+            backup_pool.submit(archive_image, result, from_api)
+        return results
+
+
+def combine_images(images, width, height):
+    if len(images) == 1:
+        return images[0]
+    vertical = width > height
+    image_width, image_height = images[0].size
+    size = (
+        (image_width, image_height * len(images))
+        if vertical
+        else (image_width * len(images), image_height)
+    )
+    combined = Image.new(images[0].mode, size)
+    for index, image in enumerate(images):
+        combined.paste(
+            image,
+            (0, index * image_height) if vertical else (index * image_width, 0),
         )
+    combined.info.update(images[0].info)
+    return combined
 
-        backup_pool.submit(archive_image, result, from_api)
-        return result
+
+def scale_image(image, scale):
+    if scale == 1:
+        return image
+    return image.resize(
+        (round(image.width * scale), round(image.height * scale)),
+        Image.Resampling.LANCZOS,
+    )
 
 
 def archive_image(image, from_api):
@@ -1991,6 +2020,7 @@ def generate(
     detailers,
     width=1152,
     height=896,
+    batch_size=DEFAULT_BATCH_SIZE,
     sampler=DEFAULT_SAMPLER,
     scheduler=DEFAULT_SCHEDULER,
     steps=DEFAULT_STEPS,
@@ -2021,6 +2051,7 @@ def generate(
         detailers=clean_detailers(detailers),
         width=width,
         height=height,
+        batch_size=batch_size,
         sampler=sampler,
         scheduler=scheduler,
         steps=steps,
@@ -2038,23 +2069,22 @@ def generate(
         denoise=denoise,
         return_scale=return_scale,
     )
-    image = generate_image(request, False)
-    if return_scale != 1:
-        image = image.resize(
-            (
-                round(image.width * return_scale),
-                round(image.height * return_scale),
-            ),
-            Image.Resampling.LANCZOS,
-        )
-    return image
+    return [
+        scale_image(image, return_scale)
+        for image in generate_images(request, False)
+    ]
 
 
 def api_generate(body, p=""):
     if not valid_pass(p):
         raise gr.Error("Invalid password")
 
-    image = generate_image(DirectRequest.model_validate_json(body))
+    request = DirectRequest.model_validate_json(body)
+    image = combine_images(
+        generate_images(request),
+        request.width,
+        request.height,
+    )
     temp = tempfile.NamedTemporaryFile(suffix=".epng", delete=False)
     temp.write(encrypted_image_bytes(image))
     temp.close()
@@ -2334,6 +2364,7 @@ def workflow_values(workflow):
         inputs.get("sampler_name", DEFAULT_SAMPLER),
         inputs.get("scheduler", DEFAULT_SCHEDULER),
         inputs.get("cfg", DEFAULT_CFG),
+        latent.get("batch_size", DEFAULT_BATCH_SIZE),
     )
 
 
@@ -2348,6 +2379,7 @@ def run_job(job_id, workflow):
             sampler,
             scheduler,
             cfg,
+            batch_size,
         ) = workflow_values(workflow)
 
         image = queued_generate(
@@ -2360,6 +2392,7 @@ def run_job(job_id, workflow):
                 sampler=sampler,
                 scheduler=scheduler,
                 cfg=cfg,
+                batch_size=batch_size,
             )
         )
 
@@ -2652,14 +2685,7 @@ def direct(
     except ValueError as error:
         raise HTTPException(422, str(error)) from error
 
-    if body.return_scale != 1:
-        image = image.resize(
-            (
-                round(image.width * body.return_scale),
-                round(image.height * body.return_scale),
-            ),
-            Image.Resampling.LANCZOS,
-        )
+    image = scale_image(image, body.return_scale)
 
     return Response(
         content=encrypted_image_bytes(image),
@@ -2993,6 +3019,13 @@ with gr.Blocks(title="Image generation") as demo:
                     with gr.Row():
                         width_input = gr.Number(1152, label="Width", precision=0)
                         height_input = gr.Number(896, label="Height", precision=0)
+                        batch_size_input = gr.Slider(
+                            1,
+                            MAX_BATCH_SIZE,
+                            DEFAULT_BATCH_SIZE,
+                            step=1,
+                            label="Images",
+                        )
                     with gr.Row():
                         sampler_input = gr.Dropdown(
                             SAMPLER_NAMES,
@@ -3090,10 +3123,11 @@ with gr.Blocks(title="Image generation") as demo:
                 )
                 button = gr.Button("Generate", variant="primary")
             with gr.Column(scale=6, min_width=420):
-                output = gr.Image(
+                output = gr.Gallery(
                     label="Preview",
                     format="png",
                     elem_id="output",
+                    columns=2,
                 )
 
     api_button = gr.Button(visible=False)
@@ -3126,6 +3160,7 @@ with gr.Blocks(title="Image generation") as demo:
             detailer_input,
             width_input,
             height_input,
+            batch_size_input,
             sampler_input,
             scheduler_input,
             steps_input,

workflow_api.py

@@ -844,7 +844,7 @@ def image_metadata(
     latent = [add("EmptyLatentImage", {
         "width": config["width"],
         "height": config["height"],
-        "batch_size": 1,
+        "batch_size": config["batch_size"],
     }), 0]
     samples = sample(
         model,