Film Grain Redmond LoRA for FLUX.2-Klein-9B

TL;DR

• LoRA adds realistic photographic film grain to FLUX.2-Klein-9B outputs
• Trigger with FILMGRAIN and FilmGrainAF in prompts
• 81 MB Safetensors file, Apache 2.0 license, works in ComfyUI/A1111/diffusers

What This Model Does

Film Grain Redmond is a style LoRA that overlays authentic film grain texture onto images generated by FLUX.2-Klein-9B (black-forest-labs/FLUX.2-klein-9B). Unlike simple noise overlays, the grain integrates naturally with image content — shadows, highlights, and midtones all receive appropriate grain density, mimicking real film stock behavior.

The model was trained by ArtificialGuyBR using GPU compute sponsored by Redmond.AI. Training data consisted of film grain style images, resulting in a LoRA that applies the effect consistently across diverse subjects — portraits, landscapes, product shots, and abstract compositions all gain the same cohesive filmic look.

How to Use

Trigger Words

Include both tokens in your prompt:

FILMGRAIN, FilmGrainAF

Recommended Settings

• Base model: FLUX.2-Klein-9B (primary) or ZImageTurbo (also compatible per Civitai)
• LoRA weight: Start at 0.8–1.0; adjust to taste
• Sampler: Euler a or DPM++ 2M Karras (standard FLUX samplers)
• Steps: 20–30
• CFG: 1.0–3.5 (FLUX typically uses low CFG)
• Resolution: 1024×1024 (training resolution)

ComfyUI Workflow

1. Load FLUX.2-Klein-9B checkpoint
2. Add LoraLoader node, select film_grain_redmond.safetensors
3. Set strength_model to 0.9, strength_clip to 0.9
4. Include trigger words in positive prompt
5. Generate

The author specifically recommends ComfyUI for best results, but the LoRA works in Automatic1111, Forge, and diffusers pipelines as well.

Example Output

The sample images show the grain rendering naturally across different lighting conditions and subject matter. Fine detail is preserved — the grain doesn't obscure textures like skin pores, fabric weave, or foliage. Instead it sits on top as a unifying texture layer, much like actual film scan.

License

Apache 2.0 — free for commercial and non-commercial use, modification, and distribution with attribution.

Common Errors and Fixes

Error | Cause | Fix | Source

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

Grain too strong/weak | LoRA weight not tuned | Adjust strength_model 0.5–1.2 | Author recommendation Trigger words not working | Wrong base model | Use FLUX.2-Klein-9B or ZImageTurbo | Civitai base model tag Color shift | LoRA trained on specific palette | Reduce strength_clip to 0.5–0.7 | Community reports

FAQ

Q: Does this work with FLUX.1-dev or FLUX.1-schnell? A: Not officially tested. The LoRA targets FLUX.2-Klein-9B architecture; results on other FLUX variants may vary.

Q: Can I combine this with other style LoRAs? A: Yes. Stack at reduced weights (0.4–0.6 each) and test. Film grain typically blends well with artistic style LoRAs.

Q: What's the difference between this and a post-process grain overlay? A: A LoRA integrates grain during generation — it adapts to local contrast, color, and luminance. Post-process overlays apply uniform noise regardless of image content.

Sources

• HuggingFace model card: https://huggingface.co/artificialguybr/FILMGRAIN-REDMOND-FLUXKLEIN9B — base model, trigger words, license, usage notes
• Civitai model page: https://civitai.com/models/233140 — download stats, base model compatibility, community metrics
• Sample images inspected directly from HF repository — visual verification of grain quality and consistency

*Related: FLUX.2-Klein-9B base model • ComfyUI LoRA loading guide • More film style LoRAs