| --- |
| license: apache-2.0 |
| base_model: black-forest-labs/FLUX.1-dev |
| tags: |
| - text-to-image |
| - flux |
| - flux-diffusers |
| - lora |
| - gguf |
| - typography |
| - high-definition |
| inference: false |
| --- |
| |
| # KielForge-HD 🚀 |
|
|
| KielForge-HD is a high-fidelity, high-definition text-to-image typography model built on top of the **FLUX.1 [dev]** architecture. This model represents a direct generation upgrade from the SDXL-based `KielForge-fast`, utilizing a significantly expanded dataset matrix to achieve clean, crisp text rendering alongside complex prompt structures. |
|
|
| The model weights are delivered using highly optimized zero-RAM binary stream layouts, providing compressed structural checkpoints suited for consumer-tier hardware runtimes. |
|
|
| ## Model Details |
|
|
| - **Developed by:** kielTech |
| - **Model Type:** Text-to-Image / Typography Fine-Tune (LoRA / GGUF Layouts) |
| - **Base Model:** black-forest-labs/FLUX.1-dev |
| - **Dataset Scale:** 2,000 High-Capacity Ingested Pairs (CSU-JPG/TextAtlas5M CleanTextSynth Subset) |
| - **Quantization Formats:** 4-bit (`Q4_0`) and 8-bit (`Q8_0`) GGUF layout streams |
| - **License:** Apache 2.0 |
|
|
| ## Intent & Usage |
|
|
| This model is explicitly tuned for text integration within high-resolution imagery. It excels at rendering legible characters, structural fonts, and custom text arrangements embedded natively within photographic or stylized scenic compositions. |
|
|
| ### Prompting Recommendations |
| To best trigger the enhanced text capabilities inherited from the underlying TextAtlas structure: |
| - Emphasize character details by using explicit layout keys like: `"displaying the text: 'YOUR TEXT HERE'"` or `"with text reading..."`. |
| - State high-fidelity target resolutions directly in the prompt layout (e.g., `"1024x1024"`, `"high-resolution photography"`). |
|
|
| ## Quantization Layout Specification |
|
|
| Tensors are processed and arranged directly into the GGUF file layout under standard GGML packing specifications: |
|
|
| | Quantization Type | GGML Type ID | Block Size | Target Use Case | Memory Profile | |
| | :--- | :--- | :--- | :--- | :--- | |
| | **Q4_0** | `2` | 32 elements | Maximum compression, ultra-low VRAM setups | Flattened Zero-RAM Stream | |
| | **Q8_0** | `8` | 32 elements | High production fidelity and text accuracy | Flattened Zero-RAM Stream | |
|
|
| ## Repository Tracking |
|
|
| The artifacts for this model run are tracked and published under the production namespace here: |
| - **Hugging Face ID:** [kiel2/KielForge-HD](https://huggingface.co/kiel2/KielForge-fast/tree/main) |
