Agent skill · marketing growth · jeremylongshore
ideogram-performance-tuning
Optimize Ideogram latency and throughput through route choice, bounded media, concurrency, async handling, and immediate storage. Use when improving an established workload without weakening quality or safety. Trigger with "speed up Ideogram", "profile Ideogram latency", or "optimize Ideogram throughput".
Why this skill is useful
Provides specific caching strategies and API configurations for optimizing Ideogram performance that the AI wouldn't reliably generate on its own.
What it needs
Requires node installed locally. Requires ideogram account access. About 2k tokens when loaded. Last updated 2026-08-07. 2,604 stars on the source repository.
What this skill does
Ideogram Performance Tuning Overview Improve end-to-end time and useful throughput with measurements, not unsupported rendering flags. Separate queue wait, upload, vendor generation, webhook or polling, download, validation, and storage so optimization preserves safety, quality, cost, and asset durability. Prerequisites Representative synthetic workload, latency and throughput SLOs, quality floor, and cost ceiling. Per-stage metrics, current endpoint mix, output count, media sizes, and concurrency policy. Approved benchmark budget, isolated destination, and cleanup owner. Current Contract Ideogram offers synchronous and asynchronous routes plus V4, transparency, P-Image, V3, and outcome-focused tools. Route-specific rendering options differ; V4 FLASH currently returns 400. Default capacity is 10 in-flight requests, and returned asset URLs require prompt download. Authentication Benchmark workers use server-side IDEOGRAMAPIKEY through Api-Key to https://api.ideogram.ai. Metrics label endpoint, status, stage, and content-free workload class, never prompt or image content. Instructions 1. Measure queue, upload, generation, reconciliation, download, validation, and storage latency separately at p50, p95, and p99. 2. Confirm the selected endpoint is the narrowest route that meets model, transparency, edit, or tool requirements. 3. Bound input bytes, dimensions, output count, and post-processing; reject work unlikely to meet its deadline. 4. Move long work to async, persist generationid, acknowledge application requests early, and reconcile by webhook plus polling. 5. Tune shared concurrency below observed capacity while tracking 429, timeout, queue age, and cost per useful output. 6. Compare one change at a time against the quality and safety floor; canary before rollout. 7. Remove benchmark assets and restore the prior setting when any guardrail regresses. Tool Discipline Use Read, Glob, and Grep to inspect metrics, queues, adapters, and fixtures. …
How to use it
Reference it in AdaL, Claude Code, Cursor or any coding agent — nothing to install:
@skills jeremylongshore/ideogram-performance-tuning--c0a603