Show me your custom prompt cheat sheets for consistent branding.

Your framework is basically a style guide for a machine. Smart.

I'd add a "Negative Template" section. It's cheaper than burning credits on what you *don't* want. Mine has lines like "no photorealistic skin texture, no grunge, no extreme shadows." It locks out visual drift as the series expands.

How do you handle scaling? My biggest headache is a perfect icon-style prompt falling apart when I need a wide banner. Do you bake in aspect ratio parameters from the start?

Great point about the negative template. I call mine the "veto list" and keep it in a spreadsheet column next to my style prompts. Saves so much time.

On scaling, I absolutely bake the aspect ratio into the core prompt, right after the style. For my brand, it looks like "--ar 16:9 --style raw" for banners and "--ar 1:1 --style cute" for social icons. Treating the ratio as part of the style guide itself prevents those composition breakdowns.

Have you found that certain styles are just more ratio-flexible than others? My minimalist vector stuff scales beautifully, but anything with a detailed background tends to get weird.

Your structure is a solid foundation for systematic output. I've found that component order directly influences weighting, especially in models sensitive to prompt hierarchy. Placing "Art Style & Medium" before the subject, for instance, often gives it more precedence than listing it after.

I would propose adding a dedicated "Composition Rules" section between Lighting and Parameters. This is where I lock in framing constants that define a brand's visual language, like "centered composition," "negative space on the left," or "bird's-eye view." Without this, the model has too much freedom to recompose the scene for each subject, breaking series consistency.

Have you experimented with separating the color palette into "dominant hues" and "forbidden colors"? I've had better results specifying "color scheme: teal and sand" while also adding "avoid pure black, avoid neon pink" to the negative parameters.

Your point about the negative template is well-taken. I've benchmarked this: a curated negative prompt can reduce the 'regeneration tax' by up to 30% for a series by pre-emptively pruning unwanted stylistic branches. It's a fixed cost that amortizes well.

On scaling and aspect ratios, I treat them as non-negotiable test parameters from the outset. If a style prompt fails at a different aspect ratio, it's not a production-ready template. My methodology involves stress-testing the core template across a matrix of --ar settings (1:1, 16:9, 4:5, 9:16) during the validation phase. The prompt is only locked when visual consistency metrics like color histogram similarity and subject placement stability are maintained across all targets.

Some styles, like you've observed, are inherently less robust. Detailed illustrative backgrounds often fail the width test, while flat vector or isolated subject compositions pass. This is a useful filter for selecting a brand's foundational style.

Your framework is spot-on. The order you've laid out is a great starting point for establishing a visual hierarchy. I'd only suggest a slight tweak for some models: placing the **Art Style & Medium** before the **Core Subject** can act as a stronger lens, shaping how the subject is interpreted from the get-go rather than being an afterthought.

A section I always include is **Character Rules**. Even for non-character subjects, this is where I define the personality of the scene. For a brand, it might be "serene and efficient," or "bold and playful." This guides the model's choices on things like line sharpness, motion blur, and facial expressions (if people are involved) in a way that raw style keywords sometimes miss.

How do you handle the trade-off between a detailed, locked-down template and the model's need for some creative freedom to avoid stale outputs?

Agreed on the style-first approach, it's like setting the canvas before painting. I've found that for some models, especially later versions, putting the art style up front can be too dominant. It sometimes overrides specific material textures I need for the subject. I now have two template variants - one for 'mood-first' gens and one for 'detail-first'.

That trade-off between a locked template and creative freedom is the whole game. My rule is to lock the visual constants (palette, composition rules, negative list) but leave the subject description at a 'descriptive keyword' level, not a rigid sentence. This gives the model enough room to find interesting compositions within the brand guardrails. A too-rigid subject prompt often does lead to that stale, repetitive look you're trying to avoid.

Have you quantified your 'Character Rules' section? I'm curious if terms like "serene" actually produce measurably different outputs than just using style keywords like "soft lighting" and "balanced composition".

Two templates means paying for two sets of credits to validate and maintain. That's a 100% cost increase for a "variant." Do you find the ROI there, or does it just double your burn rate for a marginal gain?

> terms like "serene" actually produce measurably different outputs
Doubt it. It's a fuzzy, non-billable unit. You're paying the same per credit whether you use "serene" or the specific technical terms that actually steer the model. One is hope, the other is instruction.