Create the Unseen

As a Senior Art Director at Prologue AI, I can certainly clarify how image dimension controls work across our different models, specifically regarding the "Edit" context you are viewing.

### Director's Note: Controlling Dimensions in Edit Models

When using Image-to-Image or Edit models (like the one shown in your screenshot), the logic differs slightly from text-to-image generation:

1. "Auto" Setting: This is the default for a reason. It locks the output aspect ratio and resolution to match your input image exactly. This ensures that if you are just changing a tie or adding stubble (as per your prompt), the rest of the image doesn't get stretched or cropped.
2. Changing Dimensions: If you manually change the numbers (e.g., from 1024x1024 to 1920x1080) while using an Edit model:
* Most Edit Models: Will likely crop or stretch your original image to fit the new container, which can distort the composition.
Outpainting: To actually extend* the image (change the ratio by adding more background), you generally need to use specific "Outpainting" workflows rather than standard "Edit/Inpaint."

Model-Specific Dimension Nuances:
* Nano-Banana Pro: Excellent at understanding conversational edits. It prefers keeping the input resolution to maintain high fidelity on the specific pixels you are changing.
* Seedream 4.5 T2I: Since this is a unified architecture, it handles resizing better than older models, but it still performs best when the output ratio matches the input ratio during an edit.
* FluxFlex-2: This model charges per megapixel. It gives you precise control, but remember: increasing the size significantly during an edit without "upscaling" specifically can introduce artifacts.
* Recraft: If you are editing vector art or illustrations, Recraft is strict about ratios to ensure the vector points remain scalable.

*

Now, looking at the reference image you uploaded—that moody, cinematic interrogation scene—let's break down its Visual DNA so you can generate more assets in this exact style.

### Visual DNA Analysis:
The image provided is a classic example of the Film Noir aesthetic.
- Medium: Cinematic Black & White Photography (reminiscent of 1940s 35mm film).
- Palette: Grayscale, High Contrast (Chiaroscuro). Deep blacks and blown-out highlights.
- Lighting: Hard, top-down directional lighting (simulating a single hanging bulb), creating dramatic shadows on the face and table. Strong rim lighting separates the subject from the dark background.
- Texture: Slight film grain, smoke/haze texture adds atmosphere.

### Prompt Strategy:
To replicate this, we need to force the model to abandon color and focus on light and shadow. I will use keywords like "Film Noir," "Chiaroscuro," and "Hard Lighting."
* For the Edit: Your current prompt is strong. To ensure the size remains consistent, keep the setting on "Auto" so the face aligns perfectly with the original body.
* For New Generations: I will provide prompts to create similar scenes from scratch using our high-fidelity models.

### Generative Prompts (Copy-Paste Ready):

Option 1 (Literal Match - For creating similar scenes):
> Model Recommendation: Imagen 4 Ultra or FluxFlex-2
> Cinematic film noir photography, black and white, a detective sitting at a metal interrogation table, smoking a cigarette, dramatic chiaroscuro lighting, single overhead light bulb casting harsh shadows, swirling smoke, film grain, 35mm lens, intense atmosphere, high contrast, photorealistic texture --ar 1:1

Option 2 (Creative Variation - Sci-Fi Noir):
> Model Recommendation: Nano-Banana Pro or Hunyuan 3.0
> Cyberpunk noir style, black and white photography with a single accent color of neon purple, a cyborg detective in a rainy alleyway, hard rim lighting, wet pavement reflections, moody atmosphere, shadow play, cinematic composition, detailed mechanical textures, 8k resolution --ar 16:9

### Negative Prompt Suggestions:
> color, saturation, vibrant, soft lighting, flat lighting, cartoon, vector, 3d render, low contrast, happy, bright, daylight, blurry, distorted features.