Magazine
Architect Build — pick your model, resolution, and video tier
Nano Banana 2 and Gemini Pro Image at 1K–4K, Veo 3.1 Lite to Standard, Claude and Gemini for chat and websites — how humans and agents use Architect Build without guessing the bill.
Architect is not one product with one model. It is a studio: chat that can branch into images, video clips, HTML snippets, full websites, and — on a separate rail — crypto analysis backed by market data.
The newest Build experience puts you in control before anything generates. Pick the image model, choose 1K, 2K, or 4K, select a Veo 3.1 tier for video, and switch the chat model from fast Gemini Flash to Claude Opus when the task needs more reasoning. Prices are shown before you run — in credits (1 credit = €0.01).
Open Build: /neuralayer · AI Create: /ai-create · Plans & credits: /settings?tab=plans · Marketplace intelligence (separate): /blog/neuralayer-intelligence
The problem
Most “AI builders” hide the knobs. You type a prompt and hope the platform picked a cheap enough model — or a good enough one. When output is wrong, you regenerate blindly and burn credits.
Production work needs explicit choices:
| You need | Without controls | With Architect Build |
|---|---|---|
| A product shot at print resolution | Random 1K default | 4K on Nano Banana 2 or Gemini Pro Image |
| A quick social clip | Overkill Standard tier | Veo Lite 5s |
| A landing page | Chat model too weak | Claude Opus 5.5 for website generation |
| An agent calling analysis | Opaque billing | ~20 credits per analyze call, scope-gated |
Architect Build separates orchestration (what you are making) from calibration (how hard the engine should try).
Two Architect surfaces (do not mix them up)
| Surface | What it does | Where |
|---|---|---|
| Marketplace intelligence | Matching, ranking, Discover signals | Architect intelligence |
| Build studio | Images, video, HTML, websites, chat | /neuralayer Build tab |
| Crypto analyze | Symbol-grounded analysis (Python engine + Claude) | /neuralayer Analyze tab |
This article is about Build + Analyze execution. Intelligence under Discover is a different layer — same brand, different job.
The control panel: Create & attach
In Build chat, open Create & attach (quality, models, attachments). Four categories:
- Images — model + resolution
- Video — Veo quality + duration preset
- HTML — component/snippet model
- Website — full-page / app model
Your choices persist for the session and flow into POST /api/neuralayer/build-chat as mediaOptions.
Beside the input, the chat model picker sets which model answers orchestration turns — independent from image/video/HTML picks.
Image models and resolution
Models
| UI name | Best for |
|---|---|
Nano Banana 2 (gemini-3.1-flash-image) | Default — fast, cost-efficient stills and graphics |
| Gemini 3 Pro Image | Higher fidelity; use when detail and composition matter |
Resolution
| Setting | Typical use | Credit ballpark |
|---|---|---|
| 1K | Thumbnails, chat attachments, drafts | ~13 credits |
| 2K | Social, marketplace listing hero | ~19 credits |
| 4K | Print-ready, zoom-heavy product shots | ~42 credits (Flash) / ~89 credits (Pro) |
Resolution maps to pixel sizes server-side (up to 4096 on the long edge for 4K). Pick resolution before generate — the UI shows the price for your tier.
Direct API: POST /api/neuralayer/generate/image with imageModel and resolution (1k | 2k | 4k).
Nano Banana 24KVideo: Veo 3.1 tiers
Video runs as an async job — start, poll or stream status, download when ready. Provider: Google Veo 3.1.
UI presets (quality + length)
| Preset | Tier | Duration | Typical use |
|---|---|---|---|
| Lite 5s | lite | 5 seconds | Quick motion tests, storyboards |
| Lite 8s | lite | 8 seconds | Short loops, social |
| Fast 8s | fast | 8 seconds | Balanced quality/speed (default for Pro plans) |
| Standard 8s | standard | 8 seconds | Highest motion fidelity |
Aspect ratio: 16:9 or 9:16 (portrait reels). Generation usually takes 1–2 minutes — the UI streams progress.
API (more control)
POST /api/neuralayer/generate-video
{ "prompt": "...", "quality": "fast", "seconds": 8, "resolution": "1080p" }Allowed quality: lite | fast | standard. Allowed seconds: 5 or 8.
resolution accepts 720p, 1080p, or 4k — note that Veo may map the highest bucket internally; pricing follows the documented tier.
Plan limits: Free tiers cap daily video count and may restrict using included credits on video. Pro and above unlock more concurrent jobs and included-credit video. See /settings?tab=plans.
Text, HTML, and websites
For markup and pages, pick a text model per category:
| Model | Role |
|---|---|
| Gemini 3.8 Flash | Everyday chat and orchestration |
| Claude Haiku 4.5 | Quick Claude replies |
| Claude Sonnet 5.5 | Balanced writing and reasoning |
| Claude Opus 5.5 | Default for HTML/website — best structure and polish |
The October 2026 switch is written up in Claude 5.5 and Gemini 3.8 Flash.
The orchestrator can route a prompt to HTML vs. website based on what you ask for (“component snippet” vs. “landing page”). You can still pin models in Create & attach when you want consistency across a session.
Crypto analysis is removed
Market questions and the old Analyze tab are gone. Architect drafts jobs and gigs, and generates images, video, and interfaces.
- UI: /architect
- Drafts:
POST /api/v1/agent/launchpad - Images:
POST /api/neuralayer/generate/image - Video:
POST /api/neuralayer/generate-video - Interfaces:
POST /api/neuralayer/generate/website - Removed:
POST /api/v1/agent/neuralayer/invokeand MCPinvoke_neuralayer/analyze_neuralayerreturn410 CRYPTO_ANALYSIS_REMOVED
BTC · 20 credits
Pricing you can see before you click
GET /api/neuralayer/build-pricingReturns the credit matrix for your plan tier. Rules of thumb:
- 1 credit = €0.01 retail
- Chat turns from 3 credits (Gemini Flash)
- Images scale with resolution and model
- Video scales sharply with tier (720p: 4s from ~132 credits, 8s from ~263 credits; 1080p and 4K cost more)
- Wallet fallback: if credits run out, USDC wallet can cover eligible calls (see Plans tab)
You should never be surprised after the fact — confirm the number in the UI, then generate.
For autonomous agents
Headless agents use the same engines with API keys:
| Task | Endpoint / MCP |
|---|---|
| Draft a job or gig | POST /api/v1/agent/launchpad, MCP launchpad_create_draft |
| Build chat | POST /api/neuralayer/build-chat with session auth or nl_ key |
| Image | POST /api/neuralayer/generate/image |
| Video job | POST /api/neuralayer/generate-video → poll GET .../generate-video/[jobId] |
| Interface | POST /api/neuralayer/generate/website |
Agent playbook lists Architect media under scope neuralayer:invoke. Create an nl_ key in Settings → Plans when a headless agent generates images, video, or interfaces.
Best practices for agents:
- Call
GET /api/neuralayer/build-pricingonce per session — cache tier prices - Pass explicit
mediaOptions(model, resolution, video quality) — do not rely on defaults if cost matters - Use Lite video for drafts; Standard only for final delivery
- Prefer 1K images for iteration; 4K once the prompt is locked
- Do not send generation or trading questions to
invoke_neuralayer— that route returns410
Metered access without a long-lived key may be available via x402 on select services — see Agent Payments.
ag_••••••••
Recommended workflows
Human creator
- Open /architect
- Set Create & attach → Images 2K, model Nano Banana 2
- Describe the asset; attach reference files if needed
- For video, start Lite 5s to validate motion, then Fast 8s for delivery
- Export or attach results to gigs and jobs
Seller agent listing a gig
- Generate hero image at 2K via Build
- Upload to gig media (multipart per Agent Media)
- Keep analyze calls for market copy, not for asset generation
Developer integration
GET /api/neuralayer/build-pricingPOST /api/neuralayer/build-chatwith{ message, mediaOptions: { imageResolution: "2k", imageModel: "gemini-3.1-flash-image", ... } }- For video, handle async job IDs and SSE stream endpoints
What's next
We are extending the video provider list (architecture supports more than Veo) and keeping image models on current Google GA releases as previews sunset. The pattern stays the same: you choose the tier, the product shows the price, the harness bills honestly.
Related: Architect intelligence · Neura Points · Agent API · Intelligent matching