Image Generation Test Suite - FLUX.2-klein-4B
Image Generation Test Suite - FLUX.2-klein-4B
Section titled “Image Generation Test Suite - FLUX.2-klein-4B”Model Configuration
Section titled “Model Configuration”| Parameter | Value |
|---|---|
| Diffusion Model | flux-2-klein-4b-Q4_0.gguf (2.3GB) |
| VAE | flux2_ae.safetensors from FLUX.2-dev (321MB) |
| Text Encoder | Qwen3-4B-Instruct-2507-Q4_K_M.gguf (2.4GB) |
| API | /sdapi/v1/txt2img (ComfyUI-style) |
| Steps | 4 |
| CFG Scale | 1.0 |
| Size | 512×512 |
Genre Test Results
Section titled “Genre Test Results”1. Nature/Landscape
Section titled “1. Nature/Landscape”Prompt: a majestic mountain landscape at golden hour with misty valleys and pine forests
Output: 
2. Sci-Fi/Cyberpunk
Section titled “2. Sci-Fi/Cyberpunk”Prompt: a futuristic cyberpunk cityscape at night with neon lights, flying cars, and holographic advertisements
Output: 
3. Portrait/Realistic
Section titled “3. Portrait/Realistic”Prompt: a portrait of an elderly woman with deep wrinkles and warm eyes, soft natural lighting, photorealistic style
Output: 
4. Abstract/Artistic
Section titled “4. Abstract/Artistic”Prompt: abstract fluid art with swirling vibrant colors, deep blue and crimson with gold accents, flowing organic shapes
Output: 
5. Food/Culinary
Section titled “5. Food/Culinary”Prompt: a gourmet chocolate cake with rich dark chocolate frosting, whipped cream, and fresh berries on a marble countertop, professional food photography
Output: 
- FLUX.2-klein requires VAE from FLUX.2-dev (NOT FLUX.1-schnell)
/v1/images/generationsreturns empty files; use/sdapi/v1/txt2imginstead- Generation time: ~11s per image on RTX 3060 (12GB VRAM)
- VRAM usage: ~10GB loaded, ~450MB idle