“Draw a picture of a bridge that straddles a river. There is a city skyscape in the background.The letters “HLB” are prominently displayed on the front of a building on the left hand side of the bridge and just behind it. The bridge starts in the upper right corner of the picture and comes forward at the left hand side of the picture. It is daylight. and we are approaching the bridge to enter the city.”
This picture cost 297 tokens at a cost of 1.04 cents on Openrouter.com and was the AI’s first attempt. It took perhaps 5 minutes. James, it’s all yours if you want it 
James may prefer this picture ?
“The picture is awesome, but looks a bit dark and too gothic, can you brighten it up a bit, perhaps add a mermaid sitting on the base of the closest bridge support base using a laptop on her lap ?”
This picture cost 313 tokens at a cost of 1.12 cents on Openrouter.com
Or maybe this one James ?
“perfect! is it possible to replace the existing background cityscape with likeness of the Brisbane (Australia) cityscape in the background ?. The bridge should resemble the Brisbane bridge if possible but it’s not mandatory, but everything else is perfect, don’t change the mermaid, river, trees or the HLB building”
Deepseek V4 pro, reply.
Key Brisbane landmarks:
The Story Bridge is a steel cantilever bridge, quite iconic, similar in some ways to the Jacques Cartier Bridge in Montreal. It’s a through-truss cantilever bridge.
Let me create a detailed prompt for this.
This picture cost 467 tokens at a cost of 1.32 cents on Openrouter.com.
Unable to be physically present at the HLB meets ? Join remotely via one of our Jitsi meetings on your Wireless laptop!
And join us this Thursday night (7.30 pm, 5 May 2026) for @shirbo on Automation Engines - N8N, OpenClaw and Claude Code and @matthew919 for Music Assistant and other Home Assistant Creature Comforts.
I much prefer the mermaid to @jdownie’s bush turkey!!
I understand most AIs are going token based rather than subscription. I have wasted plenty of time and money in antigravity on the default AIs on offer and so this seems like a smart solution.
Is it easy to get started with openrouter?
It’s easier than eating apple pie and ice cream!
- Go to openrouter.com and start a free account.
- Pay $10 USD in credit.
- Select the ‘Explore Models” menu
- I chose “DeepSeek: DeepSeek V4 Pro” so click on that for this exercise
- Select a provider (they host the actual AI model), I chose GMICloud
- You will see a list of models they provide
- Click on DeepSeek: DeepSeek V4 Pro again in this menu
- Click on Playground, you’ll get a terminal and can try it out
The Openrouter home page has menus for itemised usage and billing complete with graphs etc.
Now I actually use GitHub - danielmiessler/Fabric: Fabric is an open-source framework for augmenting humans using AI. It provides a modular system for solving specific problems using a crowdsourced set of AI prompts that can be used anywhere. · GitHub which was in my FreeBSD package repo and allows me to talk to the AI from my CLI directly.
[tp@fbsd15 ~]$ fabric "can you analyse the circuit in a Kicad schematic file ?"
I can analyze KiCad schematic files if you share their content! KiCad stores schematics in a readable S-expression text format (`.kicad_sch` for v6+, or older `.sch`). Once you paste the text here, I can:
- List all components with designators, values, and footprints.
- Extract nets and connectivity.
- Identify subcircuits (power, MCU, analog, connectors, etc.).
- Flag possible issues (e.g., floating nets, missing power, input/output conflicts).
- Generate a netlist summary or a simplified block description.
- Answer specific questions (e.g., “What does the reset circuit do?”).
**How to provide the file:**
Open the `.kicad_sch` or `.sch` file in a plain text editor, copy the entire content, and paste it in your message. If the file is very large, you can zip it and share a link, or paste the most relevant parts.
Alternatively, you can export a netlist or a PDF and describe the circuit – I’ll work with what you give me.
Would you like to proceed? If you have a particular question about the circuit (e.g., power supply rating, filtering, logic), include that and I’ll tailor the analysis.
If you want details on Fabric, I can show you on Thursdays online meet ?
Scrub that, I have already given a ‘Fabric’ talk here and the doc is in the index.
Thanks @techman
I had a go at Fabric and made some progress. I was interested to see that the developer, Andrew Meisser, was a guest on Intelligent Machines. He is certainly highly productive in the AI space and I plan to catch up on some of his videos about Fabric and also about AI in general.
Previously you were using z.ai. Are you still using that? They seem to only have a subscription model and that is one of the things I am trying to avoid.
I’ve also followed your guide to openrouter and I’ll see if that provides the answers to my problems at a “reasonable cost”. I can also hook it into the Perplexity based AIs which I currently pay for.
No, I was using z.ai to access GLM-5.1, but they only host GLM I think, and weren’t listed as a provider when I switched to Deepseek-R2 Pro which openrouter seem to provide via several different providers, and I get switched between them. They must be charging a lot because I’ve been billed a total of $0.48 now, including the pictures the other day 
But is is the most expensive FLOSS model.
DeepSeek V4-Pro charges $0.30 per million output tokens. Claude Opus 4.7 charges $25. GPT-5.5 charges $30. That is an 83-to-100x gap, and the model on the cheap end of it just scored 80.6% on SWE-bench Verified — 0.2 points behind Claude Opus 4.6. The weights shipped April 24 on Hugging Face under an MIT license with no commercial restrictions.
The architecture is what makes the pricing defensible rather than promotional. V4-Pro is a 1.6-trillion-parameter MoE that activates only 49 billion parameters per token, and DeepSeek cut single-token inference FLOPs to 27% of V3.2 while shrinking KV cache occupancy at 1M-token context to 10% of the previous generation. T
