Prompts library — copy-pasteable Slack messages to extend each workstream

For the operator: paste any of these into the labs Slack channel (C0B75KPL50U). Claude on the Mac mini will pick it up and run.

Identity / portrait variants

Re-run the CB LoRA matrix with different seeds

in the labs channel: re-run the CB LoRA inference matrix from runs/20260529-cb-lora-v1 with seeds [19023, 555, 999] instead of [42, 7301]. Same 3 prompts. Cost ~$0.15. Surface the new PNGs on the labs server when done.

Train a LoRA for a different identity

in the labs channel: I dropped N photos of <identity> at inputs/<identity>-fixture/. Train a flux-dev LoRA with the same hyperparams as the CB run (steps=2000, lora_rank=32, trigger TOKR), produce a 6-cell inference matrix, and judge it for identity preservation. Expected spend ~$5 training + $0.30 inference + $0.25 judge.

Add a new pose/framing prompt to the matrix

in the labs channel: extend the inference matrix in runs/<run-id> with a new prompt "<prompt text>" at seeds 42 and 7301. Don't re-train, just inference + judge. Cost ~$0.10.

Compare two LoRAs side-by-side

in the labs channel: compare LoRA <version-a> against <version-b> on the same 6-cell matrix. Stack the outputs side-by-side in a new run dir. Judge identity preservation on both. Which one wins?

Single-photo identity (open architectural question)

Probe a new single-photo vendor

in the labs channel: probe <vendor-name> for single-photo identity preservation. Use cb_4.jpg as the reference. Render the same 3 prompts (front-full, upper-front, 3q-side) and judge identity. Expected to fail like the other 6, but if it doesn't we have a new architectural path.

Garment / try-on (existing labs domain — not in active scope today but here for context)

Run a tryon sweep on a new garment category

in the labs channel: ingest the garment at inputs/<garment>/, classify it, run a tryon sweep at the standard 3 model x 2 framing matrix, and judge the outputs. Surface verdict + recommendation on the labs server.

Labs server maintenance

Restart the labs server after a code change

in the labs channel: restart the labs server. Verify it's up at localhost:8080 and the ngrok URL still resolves. Surface the URL.

Add a new prompt to this library

in the labs channel: add a new prompt to docs/prompts.md — section "<section>", title "<title>", body: "<paste here>".