Jahangirhussen Arcads Image Ad Generator
Generate Meta ad image creatives via Arcads External API.
Models Available
| Model | Best For |
|---|---|
gpt-image-2 |
Typography-heavy: UI mockups, fake chat threads, comparison tables, editorial |
nano-banana-2 |
Photoreal: lifestyle, handheld products, multi-image blending |
nano-banana-pro |
Character consistency, higher quality (costs more) |
nano-banana-edit |
Inpainting / editing existing images |
Constraints
- Aspect ratios:
1:1,16:9,9:16ONLY - Max reference images: 5 (gpt-image-2), 14 (nano-banana)
- Model is LOCKED — never switch mid-task
Workflow
- Preflight — verify
.envhasARCADS_API_KEY - Input gathering — prompt, model, aspect ratio, reference images
- Prompt rewrite — match to template library
- Credit estimate — confirm before generating
- Generate — via Python script
- Visual QA — check for text garbling, anatomy errors, brand accuracy
- Handoff — pass to meta-ad-builder for deployment
API Call
import requests, base64, os
api_key = os.getenv("ARCADS_API_KEY")
auth = base64.b64encode(f"{api_key}:".encode()).decode()
payload = {
"model": "nano-banana-2", # or "gpt-image-2"
"prompt": "Your ad prompt here",
"aspectRatio": "1:1",
"referenceImages": [] # optional: list of presigned URLs
}
resp = requests.post(
"https://external-api.arcads.ai/v2/images/generate",
json=payload,
headers={"Authorization": f"Basic {auth}"}
)
print(resp.json())
QA Checklist
- No garbled text
- No extra fingers/hands
- Brand colors correct
- Text readable at small sizes
- No merged faces/features
Regenerate up to 2x with refined prompts if issues found.