BVFinish RAW Enhancement LoRA

BVFinish is an experimental architectural-visualization enhancement LoRA from Brick Visual, trained for FLUX.2 [klein] 9B. It is designed to refine raw architectural renders into more polished images with clearer materials, more coherent lighting, balanced color, and stronger final-render definition while preserving the original scene and composition.

This is an R&D release. Results depend on the input image, prompt, seed, and workflow settings, and should be reviewed before delivery.

Model details

Property Value
Base model black-forest-labs/FLUX.2-klein-9B
Model type LoRA for image-to-image architectural enhancement
Trigger token bvfinish
Recommended strength 1.0
LoRA rank 32
Training checkpoint 12,000 steps, epoch 47
Training software AI Toolkit 0.9.3
File Klein_9B_bvfinish_v01.safetensors
File size 158.03 MiB
SHA-256 b81ece83346f519029690a388b4e3b5baf10b51511f76863e94c0cc29a63f00c

Examples

Each comparison shows the original input on the left, FLUX.2 Klein 9B with BVFinish in the center, and the base FLUX.2 Klein 9B result on the right.

Lakeside glass pavilions

Lakeside glass pavilion comparison

University building

University building comparison

Urban tower

Urban tower comparison

Fire station

Fire station comparison

Installation

Download the LoRA into your ComfyUI LoRA directory:

ComfyUI/models/loras/Klein_9B_bvfinish_v01.safetensors

Using the Hugging Face CLI:

hf download BrickMomen/raw-enhancement Klein_9B_bvfinish_v01.safetensors \
  --local-dir ComfyUI/models/loras

Recommended ComfyUI setup

  1. Load FLUX.2 [klein] 9B or its FP8 single-file version.
  2. Add LoraLoaderModelOnly and select Klein_9B_bvfinish_v01.safetensors.
  3. Set the LoRA strength to 1.0.
  4. Begin the positive prompt with the trigger token bvfinish.
  5. Use the source render as the reference image for the image-to-image pass.

The validated Brick Visual workflow uses:

Setting Value
Base checkpoint flux-2-klein-9b-fp8.safetensors
Text encoder qwen_3_8b_fp8mixed.safetensors
VAE flux2-vae.safetensors
Steps 4
Sampler Euler
Scheduler Flux2
CFG 1.0
Working resolution Approximately 1 megapixel, aligned to multiples of 32

The complete editor and API workflows are available in brick-momen-sirri/brick-comfy-workflows.

Prompt guidance

Start the prompt with bvfinish, then describe the scene and the desired finished appearance:

bvfinish, modern glass pavilion beside a lake, dusk lighting, realistic materials,
refined vegetation, natural color grading, polished architectural visualization

Useful prompt concepts include the building type, viewpoint, materials, landscape, time of day, lighting, and sky conditions. Describe the intended final image rather than the editing operation.

Intended use

BVFinish is focused on architectural exteriors and interiors, including raw renders, draft visualizations, landscape context, facade materials, glass, vegetation, and artificial lighting. It is intended for experimentation, look development, and workflow research.

Limitations

  • The model is specialized for architectural visualization and may be less reliable on unrelated subjects.
  • Strong denoise values can change geometry, signage, people, vehicles, and fine facade details.
  • Generated text and small repeated structures may be inaccurate.
  • The LoRA is experimental and may be updated as testing continues.

Files

Klein_9B_bvfinish_v01.safetensors
examples/
β”œβ”€β”€ 01-lakeside-comparison.png
β”œβ”€β”€ 02-university-comparison.png
β”œβ”€β”€ 03-tower-comparison.png
└── 04-fire-station-comparison.png

Developed by Brick Visual.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW

Model tree for BrickMomen/raw-enhancement

Adapter
(112)
this model