image-gen-9router
Skill20 starsChanged 3 months agoMITView on GitHub
Generate images via 9Router. FLUX, Imagen, DALL-E, SD.
In this repository2 other files it gives its agents.
The file
Files
Image Generation via 9Router
9Router is a local OpenAI-compatible gateway that exposes image generation
through /v1/images/generations. The same service handles LLM routing (it's
the AGY model Hermes uses) and image gen. Image gen is genuinely free via
the Cloudflare provider.
When to use this skill
- User wants to generate an image from a text prompt
- User wants AI art for an Instagram post / story / reel
- User wants to compare multiple image models (FLUX vs DALL-E vs Imagen)
- User has a working 9Router setup (NINEROUTER_URL + NINEROUTER_KEY in env)
9Router is the same service as AGY
Discovery: cat ~/.hermes/config.yaml shows
model:
base_url: http://localhost:20128/v1That localhost:20128 is 9Router. Image gen and LLM chat go through the
same gateway, but with different endpoints and different model IDs. The
key (NINEROUTER_KEY) is shared.
Endpoints
| Endpoint | Purpose |
GET /v1/models/image | List available image models (with owned_by provider) |
GET /v1/models/info?id=<model_id> | Per-model params, capabilities, size enum |
POST /v1/images/generations | Generate image(s) |
POST /v1/images/generations?response_format=binary | Generate and return raw bytes |
Auth: Authorization: Bearer $NINEROUTER_KEY (required even for local 9Router).
Discover available models
curl -s $NINEROUTER_URL/v1/models/image \
-H "Authorization: Bearer $NINEROUTER_KEY" | jq '.data[].id'Common models:
cf/@cf/black-forest-labs/flux-2-klein-4b— small, fast, free (Cloudflare)cf/@cf/black-forest-labs/flux-2-klein-9b— larger, more detail (Cloudflare)cf/@cf/black-forest-labs/flux-2-dev— development build (Cloudflare)gemini/gemini-3-pro-image-preview— Google's Imagen 3openai/dall-e-3— OpenAI DALL-E 3cf/@cf/leonardo/phoenix-1.0— Leonardo Phoenix via Cloudflare
Generate an image (Python)
import httpx, os
resp = httpx.post(
f"{os.environ['NINEROUTER_URL']}/v1/images/generations?response_format=binary",
headers={"Authorization": f"Bearer {os.environ['NINEROUTER_KEY']}"},
json={"model": "cf/@cf/black-forest-labs/flux-2-klein-4b",
"prompt": "your prompt here",
"size": "1024x1024"},
timeout=120,
)
resp.raise_for_status()
open("/tmp/out.png", "wb").write(resp.content)Or use the template:
python3 templates/generate_image.py "your prompt" -o /tmp/out.png
python3 templates/generate_image.py "your prompt" --post --caption "IG caption"Response formats
| Query / Field | Returns |
| default (URL response) | JSON with data[0].url to a hosted image |
?response_format=binary | raw image bytes |
response_format: b64_json in body | JSON with data[0].b64_json base64 string |
For posting to social media, use binary and save to disk, then upload to your CDN.
Pitfalls
- Always set
Authorizationheader even though 9Router is local. Without it, returns 401Invalid API key. The env var name isNINEROUTER_KEY, value typicallysk-...35 chars. - Some providers ignore
sizeandn.gemini(nano-banana) andhuggingfaceproviders only respectprompt; they return one image at their own size. codex(gpt-5.4-image) requires ChatGPT Plus/Pro and uses SSE streaming. Not for batch use. Zernio will return 400.- FLUX.2 Klein 4B is small but free — quality is good for stylized art, weaker for photorealistic faces. Use 9B for higher fidelity.
- Image content from Cloudflare is moderated. NSFW prompts return 400 with moderation error. Don't try to bypass.
timeout=120— first call can take 30-60s for cold model load. Set generous timeout.
Available templates
templates/generate_image.py— CLI: generate + optional IG post via Zernio. Uses FLUX.2 Klein 4B by default. Supports--listto dump available models.
Reference docs
references/upstream-skill.md— full upstreamdecolua/9routerSKILL.md for/v1/images/generationsreferences/model-catalog.md— full model list withowned_byand known params
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

