img-forge: prompt engineering guide for 5 quality tiers

[stackbilt-admin]

Summary

stackbilder.com now has a public marketing page at /img-forge (see Stackbilt-dev/stackbilt-web#54) describing the 5 quality tiers and their backing models. The marketing page is visual and sales-oriented — it doesn't cover how to prompt well for each tier.

Docs site should fill that gap with a per-tier prompt engineering guide.

Proposed structure

Suggested path: /docs/img-forge/prompting (or wherever the img-forge docs section lives in the new IA).

Sections:

1. General principles — what all 5 tiers respond to (subject, style, composition, lighting, medium, camera/lens for photoreal).
2. Per-tier guides — one subsection each:
   • Draft (SDXL) — strengths: speed, classical composition. Weaknesses: text, hands, fine detail. Best for: iteration, moodboards, "does this idea work".
   • Standard (FLUX Klein) — strengths: prompt adherence, clean geometry. Weaknesses: extreme photorealism, complex multi-subject scenes. Best for: daily-driver blog art, product mockups.
   • Premium (FLUX Dev) — strengths: coherent detail, strong style transfer. Weaknesses: slower, more expensive upstream. Best for: portfolio art, editorial illustration.
   • Ultra (Gemini 2.5) — strengths: photorealism, prompt fidelity across long prompts. Best for: photo-style hero images, realistic scene composition.
   • Ultra+ (Gemini 3.1) — strengths: best-in-class detail, typography-aware, complex multi-subject. Best for: final/hero assets where quality matters most.
3. Tier selection decision tree — "I need X, use tier Y" quick reference.
4. Common failure modes — text rendering, hands/anatomy, unintended style bleed, prompt-too-long truncation. Which tier handles each best.
5. Negative prompts / style modifiers — what's supported where.
6. Cost / quota awareness — reminder that quota is shared across tiers and Ultra/Ultra+ burn the same quota unit as Draft, so pick deliberately.

Why docs, not the product page

The product page (stackbilder.com/img-forge) is for converting anonymous visitors. The docs guide is for helping signed-up users get better output. Different audiences, different shapes.

Source material

• Model cards for each upstream provider (Stability, Black Forest Labs, Google)
• Internal prompt testing data if any exists
• Community prompt libraries (civitai for SDXL, Black Forest Labs' FLUX examples, etc.) — cite where borrowed

Cross-reference

• Stackbilt-dev/stackbilt-web#54 — marketing page migration (closing alongside this being filed)
• stackbilder.com/img-forge — the live marketing page that will link to this guide once it exists