Klein_Anime HD upscaling
Model description
license: creativeml-openrail-m tags:
- flux
- lora
- anime
- upscale
- image-to-image
- detail-enhancer base_model: wikeeyang/Flux2-Klein-9B-True-V2 instance_prompt: anime style, high-res, masterpiece
🎨 Anime HD Upscale & Detail Preservation LoRA
This LoRA is a specialized tool for high-definition upscaling of anime-style imagery. Unlike traditional upscalers that often introduce unwanted artifacts, this model focuses on enhancing resolution while strictly preserving the original painterly texture and artistic fidelity.
🚀 Key Improvements
- Unmatched Consistency: Provides significantly better structural stability and character consistency compared to SeedVR2.
- Smart Detail Management: Actively suppresses "over-detailing" and chromatic aberration (color-fringe noise) that typically plague high-ratio upscaling.
- Visual Refinement:
- Noise Reduction: Cleans up blur and compression artifacts.
- Line Art Optimization: Sharpens and smooths line work for a crisp, professional finish.
- Aesthetic Integrity: Maintains a genuine "hand-drawn" feel, avoiding the artificial "plastic/3D" look common in AI upscaling.
⚡ Online Demo & Cloud Inference
You can test this LoRA directly in your browser before downloading for local deployment.
[👉 Click Here to Try on RunningHub]
(https://www.runninghub.ai/post/2052188511582867458/?inviteCode=sdhs0trb)
🎁 Exclusive Fan Benefits:
- Instant Bonus: Register via the link above to receive 1,000 points.
- Daily Rewards: Get 100 points for every daily login.
- High-End Hardware: Experience the raw power of NVIDIA 4090 (48G VRAM) for lightning-fast inference!
🛠️ Recommended Settings (Usage Guide)
For optimal performance, we recommend the following configuration:
| Parameter | Value |
|---|---|
| Base Model | Klein 9B V2 |
| LoRA Weight | `1.0` (Adjust between 0.8-1.0 based on source) |
| Sampling Steps | `8` |
| CFG Scale | `1.0` |
| Sampler | `Euler Ancestral (Euler a)` |
💡 Pro Tips
- Workflow: Best used in an Img2Img or Tile Diffusion workflow with a low denoising strength (1) for pure upscaling, or higher if you wish to re-interpret details.
- Comparison: This model is specifically tuned to fix the "grainy" output often found in high-iteration anime renders.
🔗 Links
- Developed by: wikeeyang
- Base Model: Flux2-Klein-9B-True-V2
Trigger words
You should use 高清4K to trigger the image generation.
Download model
Download them in the Files & versions tab.