🌅 V67 Golden Hour – LoRa Model Presentation for FLUX, Krea 2, Moss-Image and Z Image Turbo.

Premise: If you decide to use my model and your image is chosen for the cover of "CIVITAI," I will give you 500 Yellow Buzz—even if used in combination with other models.

R+ images will be hidden from the gallery. - LoRA model trained with Flux, Krea 2, Moss-Image and Z Image Turbo.

Putting a Like to the V67 models is worth more than giving me Buzz (yellow/green).

📖 Concept

“V67 Golden Hour” is a multi-engine LoRA designed to capture the magic of the golden hour — warm, cinematic, and timeless. Inspired by sunrise and sunset light, it evokes dreamy, romantic, and vintage atmospheres, with a particular focus on amber tones, backlighting, and the soft textures of natural light. The name recalls a classic era (’67) and the singular quality of the “golden hour.”

What is new in this version is the engine-agnostic design. The same aesthetic has been adapted and tested across four distinct generation stacks, each with its own native strengths:

EngineWhat it brings to V67 Golden HourFLUX.1 [dev] / [pro]Maximum photorealistic fidelity, strongest prompt adherence, professional cinematic renderingKrea 2Native stylistic control, sub-15-second generations, superior color and optical phenomenaMoss-ImageFast, efficient generation with turbo LoRAs that cut steps to 4–8 while preserving textureZ Image TurboExtreme speed at CFG 1, 8–12 steps, with a remarkably “organic” (non-plastic) photographic look

🏷️ Trigger Words

Main Trigger: V67 Golden Hour

Secondary Keywords: sunset, golden hour, cinematic, warm tones, vintage, atmospheric, soft light, ethereal, romantic

Cross-model note: When stacking V67 Golden Hour with other LoRAs, the trigger retains its effect, though it becomes slightly less dominant as stack complexity increases. Keep V67 Golden Hour at the beginning of the prompt in all engines.

⚙️ Recommended Generation Parameters by Engine

The parameters below are engine-specific. Do not apply FLUX settings to turbo-distilled models — the distilled engines are designed to run at CFG 1 and will degrade if CFG is pushed higher.

1. FLUX.1 [dev] / [pro]

ParameterValueModelFLUX.1 [dev] (balanced) or [pro] (maximum quality)Steps28–32Guidance Scale3.5–5.0 (lower end for more atmospheric output)LoRA Strength0.8–1.0Formats16:9 (1344×768) · 4:5 (896×1120) · 1:1 (1024×1024)

2. Krea 2

Krea 2 is the native home of the V67 aesthetic on Civitai — a dedicated V67 Krea 2 Aesthetic component already exists on the platform, trained specifically for Krea 2. Krea 2 Turbo runs at 8 steps by default.

ParameterKrea 2 RawKrea 2 TurboSteps20–308CFG / Guidance3.0–4.01.0LoRA Weight0.81.2–2.0Samplereuler / res_multistepeuler (turbo-compatible)Schedulerbetasimple / betaNative formats1:1, 4:3, 16:9, 2.35:1, 9:16same

Community consensus for Krea 2 LoRA stacking: recommended weight is 0.8 on Raw and 1.2–2.0 on Turbo. V67 Golden Hour sits comfortably within these ranges without over-saturating the amber palette.

3. Moss-Image

Moss-Image uses turbo LoRAs (4-step or 8-step DMD variants) that are separate from the V67 style LoRA. Use one turbo LoRA at a time alongside V67 Golden Hour.

Parameter8-step DMD / DMD24-step rapid-mossSteps @ 1024²84Steps @ 1536 wide86Steps @ 2048²88CFG1.01.0SamplerEuler (flow-matching)Euler-nativeTurbo LoRA Strength1.01.0V67 LoRA Strength0.8–1.00.8–1.0

At 4 steps above 1024², images look hazy — step up to 6 or 8 for larger resolutions. LoRAs trained on Moss-Image Base work on SFT as well; train on Base, generate on SFT.

4. Z Image Turbo

Z Image Turbo is a distilled, zero-CFG model. Civitai’s official recipe states: keep cfgScale at 1 and steps at 8–12; pushing either up negates the turbo speedup without meaningfully improving quality. The NVFP4 variant confirms 10 steps as the sweet spot and notes the model excels at “unposed” photography and cinematic lighting — a natural fit for V67.

ParameterValueSteps8–12 (10 recommended)CFG Scale1.0 (no negative prompt needed)SamplerEuler a / Euler (turbo-compatible)Schedulersimple / betaLoRA Strength0.8–1.0Native resolutions1024×1024, 1536×1536

Z Image Turbo limitation: it does not support img2img or image editing — only text-to-image createImage. Plan ControlNet/img2img workflows on FLUX or Krea 2 instead.

🖼️ 10 Example Prompts with “V67 Golden Hour” (Adapted for All Engines)

Each prompt is written to work across engines. Where an engine benefits from a slight phrasing adjustment, a Per-engine note follows. For turbo models (Z Image Turbo, Krea 2 Turbo, Moss-Image turbo), the prompts are already descriptive enough — turbo engines respond best to natural-language photographic descriptions.

1. Elegant Female Portrait

V67 Golden Hour, a 22-year-old woman in an elegant silk dress, walking through a field of wildflowers at sunset, golden backlight illuminating her hair, soft bokeh, cinematic portrait, warm tones, 85mm lens, professional photography, 8k.

Per-engine: 4:5 (896×1120) on FLUX; 896×1152 or 832×1216 on Z Image Turbo; Krea 2 handles 4:5 natively. On turbo engines, drop “8k” — it adds no value at 8 steps.

2. Mountain Landscape (The Dolomites)

V67 Golden Hour, dramatic landscape of the Dolomites at sunset, mountain peaks glowing with warm alpenglow, soft clouds drifting through valleys, alpine meadow in foreground, cinematic wide shot, ultra-detailed, serene atmosphere.

Per-engine: 16:9 on FLUX (1344×768) and Krea 2 (native 16:9). Z Image Turbo is optimized for 1024² and 1536² — use 1536×1536 with a wide crop if 16:9 is essential.

3. Interior Scene (Bar with Men)

V67 Golden Hour, a rustic Italian bar interior, three men drinking espresso at the counter, warm sunset light streaming through a window, volumetric lighting, dust particles in the air, vintage decor, candid moment, film photography style, rich amber tones.

Per-engine: On FLUX, Guidance 4.5 balances realism and atmosphere. On Krea 2 Raw, CFG 3.5 preserves the volumetric dust. On Z Image Turbo, “volumetric lighting” is redundant at CFG 1 — the model’s baked-in lighting handles it; keep it only if you want to reinforce the direction.

4. Sunset on the Beach

V67 Golden Hour, a serene beach at sunset, gentle waves lapping the shore, golden and pink sky reflected in wet sand, a lone seagull walking, peaceful and tranquil, medium shot, soft focus, dreamy aesthetic.

5. Still Life with Warm Light

V67 Golden Hour, a still life of sunflowers in a vintage vase on a wooden table, golden hour light streaming through a window, warm highlights and long shadows, rustic charm, shallow depth of field, macro photography, detailed textures.

Per-engine: On Z Image Turbo NVFP4, the model’s strengths in “complex material textures” make this prompt particularly effective at 10 steps. Lock the seed and vary the vase/light direction to explore compositions.

6. Romantic Couple in the City

V67 Golden Hour, a young couple holding hands and walking through a charming European street, cobblestones glistening, sunset light painting the buildings gold, romantic atmosphere, candid shot, cinematic composition.

7. Country Road at Sunset

V67 Golden Hour, a winding dirt road through a golden wheat field, old wooden fence, sun low on the horizon casting long shadows, warm and inviting, peaceful countryside, wide angle landscape.

8. Vintage Street Photography

V67 Golden Hour, a bustling street market at golden hour, vendors and shoppers, warm light highlighting fresh produce, steam rising from food stalls, vibrant colors, street photography style, authentic moment, 35mm lens.

Per-engine: Moss-Image’s SFT turbo LoRA was trained on compact-JSON scene-graph captions and responds well to structured subject–environment relationships. The vendor/produce/steam triad maps cleanly to that training.

9. Dreamy Forest Atmosphere

V67 Golden Hour, a mystical forest with sun rays piercing through tall trees, golden light filtering through leaves, mist on the ground, ethereal and magical, fairy tale atmosphere, shallow depth of field.

10. Library with Natural Light

V67 Golden Hour, an old library with towering bookshelves, warm sunset light streaming through tall arched windows, dust motes dancing in the light beams, quiet and studious atmosphere, rich wood tones, cinematic.

Per-engine: On Krea 2, this prompt benefits from the model’s native Creativity Parameter — at higher creativity, Krea 2 adds aesthetic depth and interpretation to the light beams and wood tones without changing the subject.

💡 How to Best Use “V67 Golden Hour” Across Engines

1. Place the trigger, then match the engine’s native language. Put V67 Golden Hour first in every prompt. After that, speak the engine’s language: FLUX responds to lens and film terms (“85mm,” “film photography”); Z Image Turbo responds to natural-language photographic description (“unposed,” “weathered texture”); Krea 2 responds to mood and light direction; Moss-Image turbo responds to clear subject–environment relationships.

2. Respect the CFG divide. FLUX and Krea 2 Raw use meaningful guidance (3.5–5.0 and 3.0–4.0 respectively). Krea 2 Turbo, Z Image Turbo, and Moss-Image turbo are all CFG 1 engines — they are distilled to run without classifier-free guidance, and raising CFG costs speed without adding quality.

3. Tune LoRA weight per engine, not per taste. The same aesthetic reads differently at different weights on different stacks. Start at 0.8 on Raw/Base engines and 1.2–2.0 on Turbo engines for Krea 2; 0.8–1.0 on FLUX, Z Image Turbo, and Moss-Image. If the amber tones overpower skin or sky, reduce weight before changing the prompt.

4. Img2Img and ControlNet stay on FLUX and Krea 2. Z Image Turbo does not support img2img or editing — only text-to-image. Moss-Image’s turbo LoRAs are inference-time adapters for generation, not editing tools. For photo-to-V67 transformations (strength 0.4–0.6) or ControlNet-guided composition, use FLUX or Krea 2.

5. Lock the seed after you find a composition. Once a seed produces a strong V67 Golden Hour frame, lock it and vary the prompt slightly per engine. Each engine will interpret the same seed and prompt differently — that is the value of a multi-engine LoRA, not a bug.

6. Stack deliberately, not accidentally. V67 Golden Hour can be combined freely with other LoRAs on Krea 2 and FLUX. On Krea 2 Turbo, keep the V67 weight in the 1.2–2.0 range when stacking; on FLUX, 0.8–0.9 is a safer starting point for stacks. On Moss-Image, use only one turbo LoRA at a time alongside V67.

📌 Compatibility & Platform Notes

FLUX compatibility. Designed and tested for FLUX.1 [dev] and FLUX.1 [schnell], and compatible with FLUX.1 [pro] for professional use. FLUX LoRAs from the FLUX.1 family are not interchangeable with Krea Dev, FLUX.2 Dev, or Klein 4B/9B LoRAs — use the version of V67 Golden Hour built for the correct engine.

Krea 2 compatibility. Krea 2 Raw and Turbo are open-source (weights downloadable for local ComfyUI use), and style LoRAs for Krea 2 work alongside the official Krea 2 style LoRAs in community testing. The dedicated V67 Krea 2 Aesthetic component is the native Krea 2 build of this project and carries the same 500 Yellow Buzz cover award.

Moss-Image compatibility. Train on Moss-Image Base, generate on SFT. Turbo LoRAs are available in 4-step and 8-step DMD variants; the DMD2 variant is the creator-recommended pick for slightly more contrast and detail.

Z Image Turbo compatibility. Civitai’s official orchestration recipe defines two Z-Image variants: turbo (default, CFG 1, steps 9) and base (CFG 4, steps 20) for higher fidelity when turbo is not enough. V67 Golden Hour is tuned for the turbo path, but the aesthetic transfers to base if you need stronger prompt adherence or negative prompts.

Sampler portability. The samplers and schedulers recommended above (euler, res_multistep, beta, simple) are ComfyUI-compatible. If you use a different frontend, map cfg 1.0 correctly: ComfyUI expresses “no CFG” as 1.0, while diffusers expresses the same thing as guidance 0.0.