Prompt Engineering with Multi-Turn AI Chat

What AI Chat Does

ForgeEcho AI Chat is a multi-turn assistant for creative production. Use it to explore angles, refine prompts before image generation, debug weak outputs, and draft UGC or ad scripts—then copy the final prompt or script into AI Image or AI Voice.

Treat chat as the cheap planning layer: every reply costs 0.5 credits and is saved to your dashboard Prompt Library (original input + assistant response). Winning drafts become reusable templates for the next SKU or creative test—far cheaper than discovering structure by re-rolling 2K images.

Available Chat Models

Switch models from the dropdown in the AI Chat workspace. For prompt optimization, gemini-3.1-pro or gemini-2.5-pro usually produce the most structured output.

Model routing guide

For a given SKU prompt family, pick one primary chat model and stick with it—switching between Pro and Fast mid-project changes output structure and makes batch field swaps harder to maintain.

Agentic prompt engineering

Multi-turn chat narrows the search space step by step instead of demanding a perfect prompt in one shot. Treat it as a lightweight agent loop:

1. Plan — deliverable, channel, aspect ratio, preservation rules
2. Draft — 1–3 structured variants
3. Evaluate — you or the model checks for conflicting terms (e.g. "photorealistic" + "flat illustration")
4. Improve — change one lever only (lighting, background, tone)
5. Hand off — copy the final prompt into AI Image or AI Voice

This matters most for reasoning-guided text-to-image: models like Nano Banana 2 guess less on background, lighting, and materials when the prompt hierarchy is explicit.

Step-by-Step Workflow

1. Open AI Chat

Go to AI Chat in the app. Session context is preserved within the same conversation—follow-up messages can reference earlier drafts.

2. Clarify the deliverable

State format and constraints up front:

• Image — aspect ratio, background, hero subject, brand colors
• Voice — length, tone (warm, urgent, professional), language
• Campaign — channel (TikTok, email, product page)

Example openers:

• "I need 3 visual directions for a vitamin gummy 9:16 ad."
• "Write a 15-second voiceover script for a skincare launch—calm, trustworthy."
• "This image prompt almost works; the bottle label is blurry—what should I change?"

3. Request structured optimization

When you have a rough idea but need a production-ready prompt, ask explicitly for structure. A reliable pattern:

Optimize this into a text-to-image prompt with:
- Subject (what, where in frame)
- Style and lighting
- Background and color palette
- Constraints (keep label readable, preserve shape)
- Aspect ratio: 1:1

Rough idea: skincare bottle on clean background, premium feel, Instagram ad

Ask for 3 variants (minimal studio, lifestyle, bold campaign) and compare before generating.

4. Iterate in conversation

Ask follow-ups instead of one giant prompt. Chat excels at narrow fixes and comparisons ("cinematic vs UGC", "shorter vs more emotional").

5. Generate in the right tool

• Final image prompt → AI Image (pick model and aspect ratio there)
• Final voice script → AI Voice

See Optimize Then Generate for the full loop and quality checklist.

Example Requests

AI Image prep

• "Turn this into a product-shot prompt: white background, soft shadow, label readable."
• "Three anime-style portrait directions—keep identity, vary line art intensity."
• "Rewrite for 9:16 vertical—product in lower third, headline space in top third."

AI Voice prep

• "30-second podcast intro for a design tool—friendly, not salesy."
• "Short ad read for wireless earbuds—energetic, under 20 seconds spoken."
• "Shorten this script to 120 words; keep the hook in the first line."

Cross-tool

• "I have a hero image; suggest a matching voiceover tone and opening line."
• "This product prompt worked—adapt the same mood into a 20-second UGC voice script."

Prompt Structure Cheat Sheet

When drafting image prompts through chat, aim for this order (models parse it reliably):

1. Subject — product/person, pose, placement in frame
2. Scene & background — studio white, lifestyle kitchen, abstract gradient
3. Lighting — soft window light, rim light, high-key studio
4. Style — photorealistic, editorial, UGC handheld
5. Constraints — preserve identity, keep packaging text sharp, no extra props
6. Output — aspect ratio, intended use (Amazon main, Reels cover)

Tips

1. Name the deliverable — "Amazon main image" vs "Instagram story" changes defaults.
2. One hero subject per image prompt reduces split attention.
3. Read voice scripts aloud in chat—ask to fix lines that stumble when spoken.
4. Save winners from Prompt Library for batch work; note which model and ratio worked.

Case study: skincare SKU Instagram ad (~45 min)

Context: A serum launch needs a 4:5 Instagram feed hero and a 15-second voice hook—one operator, no studio.

Winning prompt snippet (library-ready):

Single skincare serum bottle centered, seamless white background,
soft studio key light from upper left, realistic glass reflections,
label text sharp and readable, 4:5 aspect, top 30% low-texture headline zone

Lesson: Do not jump to 4K during step 2—filter composition at 1K first, then upscale. Same single-variable rule as Optimize Then Generate.

Troubleshooting decision tree

FAQ

Does AI Chat replace AI Image or AI Voice?
No. Chat plans and refines; Image and Voice produce assets.

Where do optimized prompts go?
Each exchange is stored in Dashboard → Prompt Library with original and assistant text.

Can I chat in languages other than English?
Yes. Keep product names and SKUs consistent across locales.

Related Guides

• Optimize Then Generate
• AI Voice Workflow
• Ecommerce Image Optimization