Anime Style Redmond LoRA for FLUX.2-Klein

TL;DR

• LoRA for FLUX.2-klein-9B that produces consistent anime-style character illustrations
• Trigger words: Anime, AnimedREDAF
• Apache 2.0 license, ~81 MB SafeTensor file
• Best used with ComfyUI; works with any FLUX.2-klein workflow

What This Model Does

Anime Style Redmond is a Low-Rank Adaptation (LoRA) trained on top of FLUX.2-klein-9B (black-forest-labs/FLUX.2-klein-9B) that specializes in generating anime-style character illustrations. The training was made possible by GPU compute sponsored by Redmond.AI.

The model outputs clean, consistent anime character art with:

• Sharp linework and well-defined edges
• Expressive, large anime-style eyes
• Consistent character proportions across generations
• Soft pastel color palette typical of modern anime illustration

Trigger Words

Include both trigger words in your prompt for the style to activate:

Anime, AnimedREDAF

The HF card also lists them separately as Anime and AnimedREDAF — using both together yields the strongest effect.

How to Use

Recommended: ComfyUI

1. Load your FLUX.2-klein-9B checkpoint
2. Add the LoRA via Load LoRA node
3. Set strength (start around 0.8–1.0)
4. Include trigger words in your positive prompt

diffusers (Python)

from diffusers import FluxPipeline
import torch

pipe = FluxPipeline.from_pretrained(
    "black-forest-labs/FLUX.2-klein-9B",
    torch_dtype=torch.bfloat16
)
pipe.load_lora_weights("artificialguybr/ANIME-REDMOND-FLUXKLEIN")
pipe.to("cuda")

prompt = "Anime, AnimedREDAF, anime girl with blue hair, school uniform, soft lighting"
image = pipe(prompt, guidance_scale=3.5, num_inference_steps=28).images[0]

Automatic1111 / WebUI Forge

• Place the .safetensors file in models/Lora
• Add <lora:ANIME-REDMOND-FLUXKLEIN:1> to your prompt
• Include trigger words

Recommended Settings

Setting | Value

|---|---

Base Model | FLUX.2-klein-9B LoRA Strength | 0.8 – 1.0 Guidance Scale | 3.5 – 4.5 Steps | 20 – 30 Sampler | Euler / DPM++ 2M Resolution | 1024×1024 (native) These are starting points — FLUX models are flexible. Lower strength (0.6–0.7) blends the style more subtly with the base model's capabilities.

Model Variants

On Civitai, this LoRA is published with three base model targets:

• Flux klein 9b (primary, matches HF release)
• ZImageTurbo (ZImage ecosystem variant)
• Qwen Image (Qwen-Image base)

All share the same trigger words and anime style direction. The HF release specifically targets FLUX.2-klein-9B.

License

Apache License 2.0 — free for commercial and non-commercial use with attribution.

Common Errors and Fixes

Error | Cause | Fix

|---|---|---

Style not appearing | Missing trigger words | Add Anime, AnimedREDAF to prompt Overcooked / burned look | LoRA strength too high | Reduce to 0.6–0.7 Wrong base model | Using FLUX.1-dev or Schnell | Use FLUX.2-klein-9B checkpoint Out of VRAM | FLUX.2-klein-9B needs ~24 GB | Use quantized (NF4/8-bit) or offload to CPU

FAQ

Q: Does this work with FLUX.1-dev or FLUX.1-schnell? A: No. This LoRA was trained specifically on FLUX.2-klein-9B. Using it on other FLUX versions will produce unpredictable results.

Q: Can I merge this LoRA into the checkpoint? A: Yes, most tools (ComfyUI, Kohya-ss, etc.) support LoRA merging. Merge at 1.0 strength for a standalone anime-style checkpoint.

Q: What's the difference between the Civitai variants? A: Each variant targets a different base model (Flux klein 9b, ZImageTurbo, Qwen Image). Download the one matching your base model.

Q: Is there a VAE requirement? A: FLUX.2-klein-9B uses its own latent space — no separate VAE needed.

Sources

• HuggingFace Model Card — model card, trigger words, license, usage
• Civitai Model Page — variants, stats (2.5K downloads, 1.6K likes), base model targets

Model by ArtificialGuyBR. GPU compute sponsored by Redmond.AI.