fix: correct inaccurate download size estimates in starter models (invoke-ai#8968)

Verified model sizes against Hugging Face repositories and corrected
11 descriptions that had wrong or outdated download size estimates.

Key corrections:
- T5-XXL base encoder: ~8GB → ~9.5GB
- FLUX.2 VAE: ~335MB → ~168MB (was confused with FLUX.1 VAE)
- FLUX.1 Krea dev: ~33GB → ~29GB (uses quantized T5, not full)
- FLUX.2 Klein 4B/9B Diffusers: ~10GB/~20GB → ~16GB/~35GB
- SD3.5 Medium/Large: ~15GB/~19G → ~16GB/~28GB
- CogView4: ~29GB → ~31GB
- Z-Image Turbo: ~30.6GB → ~33GB
- FLUX.1 Kontext/Krea quantized: ~14GB → ~12GB

Author: Pfannkuchensack

invokeai/backend/model_manager/starter_models.py

@@ -71,7 +71,7 @@ class StarterModelBundle(BaseModel):
     name="t5_base_encoder",
     base=BaseModelType.Any,
     source="InvokeAI/t5-v1_1-xxl::bfloat16",
-    description="T5-XXL text encoder (used in FLUX pipelines). ~8GB",
+    description="T5-XXL text encoder (used in FLUX pipelines). ~9.5GB",
     type=ModelType.T5Encoder,
 )
 
@@ -156,39 +156,39 @@ class StarterModelBundle(BaseModel):
     name="FLUX.1 Kontext dev (quantized)",
     base=BaseModelType.Flux,
     source="https://huggingface.co/unsloth/FLUX.1-Kontext-dev-GGUF/resolve/main/flux1-kontext-dev-Q4_K_M.gguf",
-    description="FLUX.1 Kontext dev quantized (q4_k_m). Total size with dependencies: ~14GB",
+    description="FLUX.1 Kontext dev quantized (q4_k_m). Total size with dependencies: ~12GB",
     type=ModelType.Main,
     dependencies=[t5_8b_quantized_encoder, flux_vae, clip_l_encoder],
 )
 flux_krea = StarterModel(
     name="FLUX.1 Krea dev",
     base=BaseModelType.Flux,
     source="https://huggingface.co/InvokeAI/FLUX.1-Krea-dev/resolve/main/flux1-krea-dev.safetensors",
-    description="FLUX.1 Krea dev. Total size with dependencies: ~33GB",
+    description="FLUX.1 Krea dev. Total size with dependencies: ~29GB",
     type=ModelType.Main,
     dependencies=[t5_8b_quantized_encoder, flux_vae, clip_l_encoder],
 )
 flux_krea_quantized = StarterModel(
     name="FLUX.1 Krea dev (quantized)",
     base=BaseModelType.Flux,
     source="https://huggingface.co/InvokeAI/FLUX.1-Krea-dev-GGUF/resolve/main/flux1-krea-dev-Q4_K_M.gguf",
-    description="FLUX.1 Krea dev quantized (q4_k_m). Total size with dependencies: ~14GB",
+    description="FLUX.1 Krea dev quantized (q4_k_m). Total size with dependencies: ~12GB",
     type=ModelType.Main,
     dependencies=[t5_8b_quantized_encoder, flux_vae, clip_l_encoder],
 )
 sd35_medium = StarterModel(
     name="SD3.5 Medium",
     base=BaseModelType.StableDiffusion3,
     source="stabilityai/stable-diffusion-3.5-medium",
-    description="Medium SD3.5 Model: ~15GB",
+    description="Medium SD3.5 Model: ~16GB",
     type=ModelType.Main,
     dependencies=[],
 )
 sd35_large = StarterModel(
     name="SD3.5 Large",
     base=BaseModelType.StableDiffusion3,
     source="stabilityai/stable-diffusion-3.5-large",
-    description="Large SD3.5 Model: ~19G",
+    description="Large SD3.5 Model: ~28GB",
     type=ModelType.Main,
     dependencies=[],
 )
@@ -644,7 +644,7 @@ class StarterModelBundle(BaseModel):
     name="CogView4",
     base=BaseModelType.CogView4,
     source="THUDM/CogView4-6B",
-    description="The base CogView4 model (~29GB).",
+    description="The base CogView4 model (~31GB).",
     type=ModelType.Main,
 )
 # endregion
@@ -695,7 +695,7 @@ class StarterModelBundle(BaseModel):
     name="FLUX.2 VAE",
     base=BaseModelType.Flux2,
     source="black-forest-labs/FLUX.2-klein-4B::vae",
-    description="FLUX.2 VAE (16-channel, same architecture as FLUX.1 VAE). ~335MB",
+    description="FLUX.2 VAE (16-channel, same architecture as FLUX.1 VAE). ~168MB",
     type=ModelType.VAE,
 )
 
@@ -719,7 +719,7 @@ class StarterModelBundle(BaseModel):
     name="FLUX.2 Klein 4B (Diffusers)",
     base=BaseModelType.Flux2,
     source="black-forest-labs/FLUX.2-klein-4B",
-    description="FLUX.2 Klein 4B in Diffusers format - includes transformer, VAE and Qwen3 encoder. ~10GB",
+    description="FLUX.2 Klein 4B in Diffusers format - includes transformer, VAE and Qwen3 encoder. ~16GB",
     type=ModelType.Main,
 )
 
@@ -745,7 +745,7 @@ class StarterModelBundle(BaseModel):
     name="FLUX.2 Klein 9B (Diffusers)",
     base=BaseModelType.Flux2,
     source="black-forest-labs/FLUX.2-klein-9B",
-    description="FLUX.2 Klein 9B in Diffusers format - includes transformer, VAE and Qwen3 encoder. ~20GB",
+    description="FLUX.2 Klein 9B in Diffusers format - includes transformer, VAE and Qwen3 encoder. ~35GB",
     type=ModelType.Main,
 )
 
@@ -821,7 +821,7 @@ class StarterModelBundle(BaseModel):
     name="Z-Image Turbo",
     base=BaseModelType.ZImage,
     source="Tongyi-MAI/Z-Image-Turbo",
-    description="Z-Image Turbo - fast 6B parameter text-to-image model with 8 inference steps. Supports bilingual prompts (English & Chinese). ~30.6GB",
+    description="Z-Image Turbo - fast 6B parameter text-to-image model with 8 inference steps. Supports bilingual prompts (English & Chinese). ~33GB",
     type=ModelType.Main,
 )