Studio Images: add the FLUX.2-dev model family

Loading unsloth/FLUX.2-dev-GGUF failed because detect_family knew only the
Qwen3-based FLUX.2-klein, so FLUX.2-dev (the full, Mistral-based Flux2Pipeline)
resolved to nothing and the load errored. Add a flux.2-dev family: Flux2Pipeline
+ Flux2Transformer2DModel over the black-forest-labs/FLUX.2-dev base repo (gated,
reachable with an HF token), with its FLUX.2 32-channel VAE and Mistral text
encoder wired for the sd-cli path from the open Comfy-Org/flux2-dev mirror.
text-to-image only: diffusers 0.38 ships no Flux2 img2img / inpaint pipeline for
dev. Frontend gets sensible dev defaults (28 steps, guidance 4), distinct from
klein's turbo defaults. Verified live: GGUF load resolves the family + gated base
repo and generates a real 1024x1024 image on GPU.

studio/backend/core/inference/diffusion_families.py

@@ -138,6 +138,24 @@ _FAMILIES: tuple[DiffusionFamily, ...] = (
             ("Comfy-Org/z_image_turbo", "split_files/text_encoders/qwen_3_4b.safetensors", "llm"),
         ),
     ),
+    # FLUX.2-dev is the full (non-distilled) FLUX.2. It uses the Mistral-based
+    # Flux2Pipeline, distinct from klein's Qwen3-based Flux2KleinPipeline, so it needs
+    # its own entry. Its base diffusers repo is gated (gated=auto) but reachable with an
+    # HF token. text-to-image only: diffusers 0.38 ships no Flux2 img2img / inpaint
+    # pipeline for dev. VAE + Mistral text encoder come from the open Comfy-Org/flux2-dev
+    # mirror for the sd-cli path (shares the FLUX.2 32-channel AE with klein).
+    DiffusionFamily(
+        name = "flux.2-dev",
+        pipeline_class = "Flux2Pipeline",
+        transformer_class = "Flux2Transformer2DModel",
+        base_repo = "black-forest-labs/FLUX.2-dev",
+        aliases = ("flux2-dev", "flux2dev"),
+        sd_cpp_vae = ("Comfy-Org/flux2-dev", "split_files/vae/flux2-vae.safetensors"),
+        sd_cpp_vae_format = "flux2",
+        sd_cpp_text_encoders = (
+            ("Comfy-Org/flux2-dev", "split_files/text_encoders/mistral_3_small_flux2_bf16.safetensors", "llm"),
+        ),
+    ),
     DiffusionFamily(
         # Instruction editing with FLUX. FluxKontextPipeline takes an input image + an edit
         # instruction; the GGUF transformer is the standard FluxTransformer2DModel, with the

studio/backend/core/inference/sd_cpp_args.py

@@ -40,6 +40,7 @@ from core.inference.diffusion_memory import (
 _TE_FLAGS_BY_FAMILY: dict[str, tuple[str, ...]] = {
     "z-image": ("--llm",),
     "flux.2-klein": ("--llm",),
+    "flux.2-dev": ("--llm",),
     "qwen-image": ("--qwen2vl",),
     "flux.1": ("--clip_l", "--t5xxl"),
 }

studio/backend/tests/test_diffusion_backend.py

@@ -45,8 +45,13 @@ def test_detect_family_from_repo_id():
     assert klein.cfg_kwarg == "guidance_scale"
     # Both klein sizes share the one family (base repo resolved per-variant).
     assert detect_family("unsloth/FLUX.2-klein-9B-GGUF").name == "flux.2-klein"
-    # Only klein is wired up; the Mistral-based FLUX.2-dev base repo is gated.
-    assert detect_family("unsloth/FLUX.2-dev-GGUF") is None
+    # FLUX.2-dev is the Mistral-based Flux2Pipeline, a distinct family from klein; its
+    # gated base repo is reachable with an HF token. It must not collide with klein.
+    dev = detect_family("unsloth/FLUX.2-dev-GGUF")
+    assert dev.name == "flux.2-dev"
+    assert dev.pipeline_class == "Flux2Pipeline"
+    assert dev.base_repo == "black-forest-labs/FLUX.2-dev"
+    assert detect_family("black-forest-labs/FLUX.2-dev").name == "flux.2-dev"
     # Qwen-Image guides via true_cfg_scale, not guidance_scale.
     assert detect_family("unsloth/Qwen-Image-2512-GGUF").cfg_kwarg == "true_cfg_scale"
     assert detect_family("unsloth/Z-Image-GGUF").cfg_kwarg == "guidance_scale"

studio/frontend/src/features/images/images-page.tsx

@@ -194,6 +194,8 @@ const MODEL_DEFAULTS: Array<{ match: string; steps: number; guidance: number }>
   { match: "kontext", steps: 28, guidance: 2.5 },
   { match: "flux.1", steps: 28, guidance: 3.5 },
   { match: "flux.2-klein", steps: 4, guidance: 0 },
+  // FLUX.2-dev is the full (non-distilled) model: more steps + real guidance, unlike klein.
+  { match: "flux.2-dev", steps: 28, guidance: 4 },
   { match: "qwen-image", steps: 20, guidance: 4 },
   { match: "z-image", steps: 20, guidance: 4 },
 ];