feat: v0.2.0 — multi-variant FIBO Generate, mask/spill presets, premultiply flow

README.md

@@ -50,13 +50,15 @@ That's it — you're ready to use the tools.
 | Erase | Remove objects with inpainting |
 | Expand | Extend image canvas with AI generation |
 | FIBO Edit | Prompt-driven image editing |
-| FIBO Edit Recipes | FIBO Edit with 99 curated preset prompts across 11 categories |
-| FIBO Generate | Text-to-image generation |
+| FIBO Edit Recipes | FIBO Edit with 100+ curated preset prompts across 11 categories — including Create Mask, Create Soft Mask, and Remove Greenscreen Spill compositing presets |
+| FIBO Generate | Text-to-image generation with multi-variant batch (1–4 in parallel) |
 | Generate VGL | Structured-prompt JSON generator (chains into FIBO nodes) |
 | Sequence Output | Batch render a chain of upstream Bria nodes across a frame range |
 
 All tools are accessible from the **Bria** menu or by pressing Tab and typing "bria".
 
+Every Bria node also includes **View Generation Data** and **Copy Generation Data** buttons (Settings tab) for inspecting or copying the most recent run's prompt, seed, response, and full sidecar JSON.
+
 ## Shared configuration
 
 Bria tools store configuration in `~/.bria/bria.json`. This file is shared across
@@ -76,10 +78,10 @@ Dev launcher scripts are also provided in `tools/dev/`.
 
 ## Documentation
 
-- Detailed setup and node reference: [docs/bria_nuke/README.md](docs/bria_nuke/README.md)
-- Configuration reference: [docs/CONFIG.md](docs/CONFIG.md)
-- Repo structure: [docs/contributing/STRUCTURE.md](docs/contributing/STRUCTURE.md)
-- Troubleshooting: [docs/bria_nuke/TROUBLESHOOTING.md](docs/bria_nuke/TROUBLESHOOTING.md)
+- Detailed setup and node reference: [documentation/bria_nuke/README.md](documentation/bria_nuke/README.md)
+- Configuration reference: [documentation/CONFIG.md](documentation/CONFIG.md)
+- Repo structure: [STRUCTURE.md](STRUCTURE.md)
+- Troubleshooting: [documentation/bria_nuke/TROUBLESHOOTING.md](documentation/bria_nuke/TROUBLESHOOTING.md)
 
 ## License
 

bria_core/utils.py

@@ -116,28 +116,61 @@ def resolve_proxies(
 
 
 def extract_image_url(data: dict) -> str | None:
+    urls = extract_image_urls(data)
+    return urls[0] if urls else None
+
+
+def extract_image_urls(data: dict) -> list[str]:
+    """Return every image URL the API response contains, in order.
+
+    Handles a few shapes Bria's endpoints have used:
+      - {"result": {"image_url": "..."}}                       (single, FIBO Generate sync)
+      - {"result": {"images": ["url1", "url2"]}}              (multi as list of strings)
+      - {"result": {"images": [{"image_url": "..."}, ...]}}   (multi as list of dicts)
+      - {"result": [{"image_url": "..."}, ...]}               (result as list of dicts)
+      - {"image_url": "..."}                                    (top-level)
+      - {"images": [...]}                                       (top-level array)
+    """
     if not isinstance(data, dict):
-        return None
+        return []
+
+    seen: set[str] = set()
+    out: list[str] = []
+
+    def _add(value: object) -> None:
+        if isinstance(value, str):
+            v = value.strip()
+            if v and v not in seen:
+                seen.add(v)
+                out.append(v)
+
+    def _walk_image_dict(obj: object) -> None:
+        if isinstance(obj, dict):
+            for key in ("image_url", "url", "imageUrl"):
+                _add(obj.get(key))
+
+    def _walk_image_list(items: object) -> None:
+        if isinstance(items, list):
+            for item in items:
+                if isinstance(item, str):
+                    _add(item)
+                elif isinstance(item, dict):
+                    _walk_image_dict(item)
+
     result = data.get("result")
     if isinstance(result, dict):
-        url = result.get("image_url")
-        if isinstance(url, str) and url.strip():
-            return url.strip()
-
-        images = result.get("images")
-        if isinstance(images, list) and images:
-            first = images[0]
-            if isinstance(first, str) and first.strip():
-                return first.strip()
-            if isinstance(first, dict):
-                first_url = first.get("image_url")
-                if isinstance(first_url, str) and first_url.strip():
-                    return first_url.strip()
-
-    url_top = data.get("image_url")
-    if isinstance(url_top, str) and url_top.strip():
-        return url_top.strip()
-    return None
+        _walk_image_dict(result)
+        _walk_image_list(result.get("images"))
+        _walk_image_list(result.get("results"))
+    elif isinstance(result, list):
+        _walk_image_list(result)
+
+    # Top-level fallbacks
+    _walk_image_dict(data)
+    _walk_image_list(data.get("images"))
+    _walk_image_list(data.get("results"))
+
+    return out
 
 
 def png_ihdr_info(path: str) -> dict | None:

docs/bria_nuke/README.md

@@ -105,7 +105,7 @@ Current node set (11):
 Enhance defaults and controls:
 
 - `preserve_alpha` defaults to **on**
-- `sync` is hidden on current new nodes and forced to `true` at runtime because async polling is not implemented for Enhance yet
+- `sync` is hidden on current new nodes and forced to `false` at runtime — Bria's `/v2/image/edit/enhance` endpoint stopped accepting `sync=true` (returns 422) in May 2026, and `BriaClient._poll_status` handles the async response transparently
 - `seed` is optional
 - `Use Mask` toggles region-limited internal composite (`Keymix`) between original and enhanced result
 - mask inputs without alpha are normalized internally (mask red -> alpha) before compositing
@@ -153,6 +153,13 @@ FIBO Generate UX highlights:
 - includes `status` field (`idle` / `processing` / `done` / `error`)
 - includes `Open Bria Dashboard` button
 - `Before` / `After` compares previous generated result against latest generated result
+- **Multi-variant batch (Generate Multiple Images)**:
+	- `Generate Multiple Images` toggle reveals a `Num Images` dropdown (2, 3, or 4), an `Image Index` slider, and a `Use This Image` button
+	- toggle ON → on Run, the node fires N parallel async calls to `/v2/image/generate` (the endpoint returns one image per call, so true batching is parallel-call style; per-variant seed offset is applied when a base seed is set so variants are actually different)
+	- saved files are suffixed `_v01.png` … `_vNN.png` next to the canonical `result_path`; the canonical path points at variant 1 by default
+	- after a multi-variant run, drag the `Image Index` slider to switch the internal Read between variants without re-running
+	- `Use This Image` promotes the currently-displayed variant to be the canonical `result_path` so downstream nodes (Sequence Output, etc.) consume the locked-in variant
+	- sidecar JSON records every variant in `result_paths`, with `result_path` / `image_path` still pointing at the first for backward compatibility
 
 FIBO Edit UX highlights:
 
@@ -197,10 +204,18 @@ Generate VGL node history:
 
 FIBO Edit Recipes UX highlights:
 
-- categorized preset launcher built on FIBO Edit's execution path; ships with 99 prompts across 11 categories (Style, Weather, Seasons, Time of Day, Camera, Lighting, Compositing, Cleanup, AI Corrections, Object Edits, plus Custom for user prompts)
+- categorized preset launcher built on FIBO Edit's execution path; ships with 100+ prompts across 11 categories (Style, Weather, Seasons, Time of Day, Camera, Lighting, Compositing, Cleanup, AI Corrections, Object Edits, plus Custom for user prompts)
 - pick a category → the matching Preset dropdown appears with that category's options; selecting a preset auto-fills the prompt field
-- **Object-targeted presets** (5 of them: delete, replace, isolate, change color, change material) reveal 5 fixed Object slots — fill in the names of objects in your image and the prompt is rebuilt with constraint-injection ("STRICT LOCKS") wording so the model only edits what you specify
-- **Compositing-targeted presets** (7 of them: harmonize lighting, match shadows, blend edges, color harmonize, fix reflections, integrate elements, match ambient) reveal 5 fixed Element slots — describe which composited elements need attention
+- **Object-targeted presets** (delete, replace, isolate, change color, change material) reveal 5 fixed Object slots — fill in the names of objects in your image and the prompt is rebuilt with constraint-injection ("STRICT LOCKS") wording so the model only edits what you specify. Isolate Object and Delete Object work with object name only (no modifier required); other presets fall back to the generic preset when the modifier is missing.
+- **Compositing-targeted presets** (harmonize lighting, match shadows, blend edges, color harmonize, fix reflections, integrate elements, match ambient, **Create Mask**, **Create Soft Mask**, **Remove Greenscreen Spill**) reveal 5 fixed Element slots — describe which composited elements need attention
+- **Create Mask** outputs a clean binary alpha matte (white over named element(s), black elsewhere) suitable as a luminance key
+- **Create Soft Mask** outputs a grayscale density matte for atmospherics — smoke, fire, dust, fog, mist — where solid objects in the same frame remain pure black
+- **Remove Greenscreen Spill** removes green/blue-screen color contamination, fringing, and edge spill from named element(s) while preserving naturally-occurring greens (foliage, signage, wardrobe). Designed to run after the keyed subject is composited over its new background.
+- **Premultiply with Mask toggle + Mask Edge Softness slider** appear automatically when Create Mask or Create Soft Mask is the active preset
+	- toggle ON → routes the API-returned mask through an internal `Blur → FilterErode → Copy(rgb=input, a=mask.red) → Premult` chain, producing a clean cutout element with anti-aliased edges
+	- toggle OFF → output is the raw API mask (the existing behavior)
+	- Mask Edge Softness drives the Blur size and a half-size FilterErode that compensates for blur expansion, so the alpha stays tight to the named element
+	- both controls are hidden for any other preset and auto-clear when the user switches preset/category, so subsequent runs do not unexpectedly produce cutouts
 - auto-vs-manual prompt tracking: if you edit the prompt manually after a preset is loaded, switching presets won't clobber your edits
 - delegates the actual API call to the existing FIBO Edit pipeline — same history, presets, metadata sidecar, and viewer behavior
 

nuke/adapter.py

@@ -468,9 +468,13 @@ def upscale_from_files(
         session=session,
     )
 
+    # As of 2026-05, /v2/image/edit/enhance returns 422 for sync=true ("sync request is not
+    # supported with this endpoint"). BriaClient._poll_status handles async responses, so
+    # default Enhance to sync=False. Other modes (increase_resolution) retain the prior default.
+    default_sync = False if endpoint_name == "enhance" else True
     payload: Dict[str, object] = {
         "image": file_to_base64(image_path),
-        "sync": True if sync is None else bool(sync),
+        "sync": default_sync if sync is None else bool(sync),
     }
 
     if seed is not None: