Multi-Model AI Workspace — Chat, Text-to-Image & TTS
ForgeEcho docs for overseas ecommerce and marketing teams—prompt engineering, Amazon/Shopify listing photos with Nano Banana 2, UGC ad batch creatives, and TTS voiceovers in one Chat · Image · Voice pipeline.
Documentation Overview
ForgeEcho is a multi-model AI creative workspace for teams that ship assets every week: AI Chat for prompt engineering and script drafting, AI Image (Nano Banana 2 / Flux / GPT Image 2) for text-to-image and reference editing, and AI Voice for TTS narration. Chat → image → voice share one credit system so Amazon/Shopify listing work and Meta/TikTok creative tests stay on the same loop—not three disconnected apps.
The underlying loop is Plan → Evaluate → Improve: clarify intent in chat, score outputs against a short rubric, then promote winners into your Prompt Library. That discipline matters when you need consistent brand visuals, batch creative testing, or weekly social output—not a single lucky generation.
Why not ChatGPT + Midjourney + a TTS tab?
| Approach | What breaks in production |
|---|---|
| Separate ChatGPT / Midjourney / ElevenLabs accounts | Prompt history fragments; hard to hand off a “winner” to the next SKU |
| Discord-only image tools | Weak for reference-locked product photography and credit forecasting |
| One-shot prompting in any model | High re-roll cost; no shared evaluation rubric across the team |
ForgeEcho keeps the same Plan–Evaluate–Improve discipline documented below, with credits and Prompt Library shared across Chat, Image, and Voice.
Who These Guides Are For
- Solo creators publishing to TikTok, Reels, podcasts, and YouTube
- Ecommerce teams refreshing listing photos and ad creatives at SKU scale
- Marketing operators running voiceovers, poster campaigns, and social visuals
- Anyone who wants a reliable chat → generate → iterate loop instead of one-off prompting
What You'll Learn
| Topic | Outcome |
|---|---|
| AI Chat | Turn vague ideas into structured image prompts and voice scripts |
| AI Image | Generate or edit visuals for products, ads, and creative assets |
| AI Voice | Produce ad reads, explainers, and social hooks from text |
| Style Transfer | Shift to anime, cyberpunk, or cinematic looks while keeping composition |
| Ecommerce Images | Listing-ready photos with accurate color and clean backgrounds |
| Portrait Retouch | Natural beauty enhancement without plastic skin |
| Batch Social | Template-driven weekly content with consistent brand look |
| Faceless Shorts | Chat → 9:16 stills → TTS → CapCut assembly |
| Credit & Models | Budgets and when to use fast / 2K / 4K |
| Brand Visual System | Locked prompt blocks + Prompt Library naming |
| Image Troubleshooting | Symptom → one-lever fixes for common failures |
| Poster Design | Campaign layouts with clear typography zones |
| World Cup AI Creative | Fan avatars, kicking poses, match commentary, memes |
| Optimize → Generate | The core loop that connects Chat, Image, and Voice |
Recommended Starting Path
- AI Chat Guide — clarify deliverables and draft prompts before generating
- Optimize Then Generate — adopt the five-step pipeline used across all scenarios
- Open the use-case guide that matches your next deliverable
AI Chat Guide
Multi-turn assistant for brainstorming, prompt drafting, and creative feedback.
How to Use AI Voice
Text-to-speech scripts and workflows for ads, explainers, and social audio.
How to Change Image Style with AI
Style transfer from reference image to finished asset.
Ecommerce Product Image Optimization
Listing photos, backgrounds, and conversion-ready variants with AI Image.
AI Photo Retouch and Beauty
Identity-preserving portrait retouch with AI Image.
Optimize Then Generate
The repeatable pipeline for Chat, Image, and Voice.
AI Poster Design Workflow
Posters and campaign covers with reusable prompts.
Social Media Batch Creative
Batch creatives for feed and short-form channels.
Faceless Short Video Pipeline
Scripts, 9:16 stills, TTS, and editor assembly for Shorts/Reels/TikTok.
Credit Budget & Model Selection
Unit costs, weekly budgets, and when to use fast vs 2K vs 4K.
Brand Visual System
Prompt Library naming and locked blocks for consistent brand look.
AI Image Troubleshooting
Fix shape drift, garbled labels, plastic skin, and bad crops fast.
World Cup AI Creative
Fan avatars in national kits, kicking poses, AI match commentary, posters, and memes.
How the Guides Connect
Every scenario guide assumes the same underlying workflow: draft in AI Chat (0.5 credits per reply), refine until the prompt or script is specific enough, generate in AI Image (3–8 credits by resolution) or AI Voice (1 credit per 500 characters), then iterate on one variable at a time. Optimized prompts are saved automatically to your dashboard Prompt Library for reuse across batch runs.
If you are new to generative tools, start with Optimize Then Generate—it explains why vague prompts fail and how structured instructions improve stability across Nano Banana 2, Flux-family models, and TTS voices.
Workspace Quick Reference
| Module | Typical use | Credits (approx.) |
|---|---|---|
| AI Chat | Prompt engineering, UGC scripts, failure diagnosis | 0.5 per reply |
| AI Image | Text-to-image, reference edit, 4K listing assets | 3–8 per image by resolution |
| AI Voice | Ad reads, short-form hooks, explainers | 1 per 500 characters |
The 2026 multi-model creative stack is intentional: Chat for prompt engineering, Nano Banana 2 for text-to-image, and TTS for narration. Chat models (Gemini family) and image models (Nano Banana 2 family) play different roles—chat interprets and rewrites; image executes pixels. Swapping those roles usually costs more credits than following the pipeline.
30-minute quick start
If you only have half an hour, run this minimal loop once:
- Open AI Chat guide and generate one structured image prompt from the sample request (~5 min)
- Paste into AI Image, pick
nano-banana-fast+ 1K, generate 2 variants (~10 min) - Score with the checklist in Optimize Then Generate, change one variable, regenerate once (~10 min)
- Save the winning prompt to your library with aspect ratio and model notes (~5 min)
After one pass, open the scenario card above that matches your next deliverable.
Pick the right guide (by deliverable)
| Your next task | Start here | Why |
|---|---|---|
| Listing photos, white background, color accuracy | Ecommerce Image Optimization | Platform specs, A/B testing, batch SKU workflow |
| LinkedIn headshot, dating/social portrait | Photo Retouch and Beauty | Identity-preserving two-pass retouch |
| Anime, cyberpunk, cinematic restyle | Image Style Transfer | Reference upload + preservation blocks |
| Event promo, sale banner, campaign cover | AI Poster Design Workflow | Typography safe zones and ratio exports |
| Weekly feed/stories at volume | Social Media Batch Creative | Templates, credit planning, fatigue testing |
| Faceless Shorts / Reels / TikTok | Faceless Short Video Pipeline | Chat → stills → TTS → CapCut |
| Credits melting / wrong model choice | Credit Budget & Model Selection | Budgets and stage-based models |
| Brand look drifts across operators | Brand Visual System | Locked blocks + library naming |
| Bad outputs, unsure what to change | AI Image Troubleshooting | Symptom → one-lever fixes |
| Ad read, explainer, social hook audio | AI Voice Workflow | Script structure, pacing, voice selection |
| Prompts feel vague or outputs drift | Optimize Then Generate | Plan–Evaluate–Improve loop before any scenario |
Model selection at a glance
| Goal | Chat model | Image model | Resolution |
|---|---|---|---|
| Quick prompt draft | Any Gemini / chat model | — | — |
| Fast visual filter (6+ variants) | — | nano-banana-fast | 1K |
| Final listing / ad still | — | nano-banana-2 | 2K |
| Print or large display | — | nano-banana-2 or nano-banana-pro | 4K |
| Voice tone test | — | — | Short clip first |
Chat interprets intent; image models render pixels. Run at least one chat round before spending image credits on a new scenario.
When outputs fail (quick diagnosis)
| Symptom | Likely cause | Fix (one lever) |
|---|---|---|
| Wrong colors / plastic skin | Over-aggressive prompt or no reference | Upload reference; reduce beauty words; split lighting vs skin passes |
| Composition breaks at 9:16 | Prompt composed for landscape | Rewrite prompt for vertical safe zones |
| Face doesn't match photo | Text-only generation | Upload reference; add "preserve identity, same person" |
| Garbled text on product | Model inventing labels | Simplify label expectations; overlay text in design tool |
| Robotic voice read | Script written for reading, not speaking | Re-draft in AI Chat with "spoken phrasing, short sentences" |
| Inconsistent batch look | Prompt drift between runs | Save winner to Prompt Library; change one variable only |
For the full evaluation rubric, see Optimize Then Generate.