---
title: Prompt Architect
category: product
entity_type: skill
price: $15
canonical: https://forgehouse.ai/skills/prompt-architect/
lang: en
hreflang_alt: https://forgehouse.ai/tr/skiller/prompt-architect/
last_updated: 2026-06-20
---

# Prompt Architect

> AI image and video prompt builder that guides users from a rough vision to a copy-paste ready…

A structured assistant that takes you from a rough visual idea to a copy-paste ready prompt for AI image and video generation through a clean three-step flow: free-form vision input, prompt generation with gap detection, and a focused iteration loop. It never generates images itself; it builds and refines the prompt, optimized by default for Nano Banana Pro images and Kling video. When inputs conflict or are over-specified, it flags the issue precisely and offers concrete next directions.

## Use cases
- Turning a rough visual idea into a usable prompt
- Writing image prompts for Nano Banana Pro
- Writing motion prompts for Kling video
- Diagnosing why a generated result felt off
- Resolving conflicting or over-specified prompt instructions
- Iterating prompts without rewriting from scratch

## Benefits
- Go from vague vision to a ready-to-paste prompt fast
- Spot missing layers before wasting generation credits
- Fix the exact prompt layer responsible for a bad result
- Keep prompts tight where confirmed, open where intentional

## What’s included
- Three-step flow: vision input, generation plus gap detection, iteration
- Clearly-defined summary separating confirmed from inferred details
- Prompt layer checklist (subject, environment, camera, lighting, lens, style, motion)
- Optional specification suggestions with concrete example directions
- Error-state labels for CONFLICT, AMBIGUOUS, and OVER-SPECIFIED inputs
- Restart command that resets the flow with no prior context

## Who it’s for
Creators and marketers who want professional AI image and video prompts without learning prompt syntax by trial and error.

## How it runs
Describe the image in your own words and skip the twenty-question intake. The skill confirms what is defined, drafts a model-ready prompt, then flags gaps and conflicts before a render burns credits.
1. Opens with free vision input: no structured questionnaire, the user describes what they imagine in their own words, optionally with a reference image for subject or style. Twenty upfront questions kill momentum, so step one asks none.
2. Summarizes what is actually confirmed under a 'clearly defined' header, listing only what was explicitly stated or can be confidently inferred, never speculating layers into existence.
3. Generates a clean, copy-paste ready prompt block optimized for the target model (Nano Banana Pro for images, Kling for video by default), written tight: specific where confirmed, intentionally broad where the user left room.
4. Runs gap detection across the seven prompt layers (subject, environment, camera angle, lighting, lens feel, style and era, motion for video) and surfaces only the missing or vague ones as optional suggestions with 1 or 2 concrete example directions each, not open-ended questions.
5. Flags input problems in neutral system language before any paid render burns credits: [CONFLICT] when two cues fight each other (cinematic lighting versus natural daylight), [AMBIGUOUS] when a layer can be read two ways, [OVER-SPECIFIED] when detail overload will distort the output.
6. Iterates on feedback surgically: diagnoses which single layer caused the complaint, explains the reasoning in 1 or 2 sentences, updates only that layer and returns the full revised prompt. Rewriting from scratch is reserved for a fundamental change of direction.

## FAQ
### I use a model other than Nano Banana Pro or Kling, does it still help?
Yes. It optimizes for those two by default, but you can specify another model and the three-step flow plus the prompt layer checklist (subject, environment, camera, lighting, lens, style, motion) carry over. The layers are model-agnostic craft.

### How does it improve a prompt beyond just rephrasing what I said?
It runs gap detection against the layer checklist, separates confirmed details from inferred ones in a structured summary, and flags CONFLICT, AMBIGUOUS, or OVER-SPECIFIED inputs with concrete next directions. The iteration loop then fixes the exact layer behind a bad result instead of rewriting from scratch.

### Will it generate the actual image or video for me?
No, it never generates media itself. It builds and refines the copy-paste ready prompt; you take that prompt to your image or video tool and spend your generation credits there.

## Price
$15, one-time, no subscription. VAT included.

Related guide: [AI and LLM engineering](https://forgehouse.ai/guides/ai-llm-engineering/)