blog: painting in the Spark's spare memory

[AdamFerguson]

New post: Painting in the Spark's spare memory — companion to the ComfyUI feature landing in spark-lab (PR #2).

• Frontmatter matches the existing schema (title/description/date/tags/category); slug spark-image-generation.
• Voice follows the LLM-lab post: problem-first, no hype, honest scope (written before the first render — the profile-picture line is deliberately conditional).
• Privacy-checked: no hostnames, private IPs, usernames, or tokens.

Worth a skim for tone before it ships — it mentions the site palette extraction and the woodcut/flat-vector plan, so if you'd rather keep profile-pic experiments off the blog I can cut the last section.

[AdamFerguson]

Rewritten per feedback:

• ~40% shorter, em-dash/parallelism/rule-of-three patterns stripped, headers say what's in them.
• New spine: three streams steered from a phone over Matrix — recipe staging, the spark-lab PR, this post — with the checkpoints that mattered (approving risky steps, killing an un-consented experiment, queuing work against "free the box tonight").
• Honest failure section: the agent confidently got Matrix E2EE wrong and the fix was reading the source. Argues the actual lesson: when the agent says "impossible," check what it read.

Still no hostnames/IPs/usernames. The profile-picture thread kept but hedged (first render genuinely hasn't run — test plan in spark-lab#2 says so).