The prompt optimizer
for every AI.

Say it rough, send it refined: type what you want in plain words and IterationFlow rewrites it the way your model actually wants it, so the work that comes back is better. Text, image and video.

Iterate better with Claude.ChatGPT.Gemini.Claude Code.Midjourney.Nano Banana.your context.every model.

The same words land differently on every model.

Rewritten the same prompt three times and it’s still not right? The words weren’t wrong. They were tuned for a different model. Here is the whole path a prompt takes.

  1. Your rough prompt

    Type it or say it. No structure, no formatting, no second pass.

  2. The model you are about to use

    Ten text targets, eight image generators and ten video models. Reasoning models and standard ones are separate targets, because they want different things.

  3. Optional

    Your own context, if you want it

    Choose a Project or a Business folder from your vault. Those notes are read at the moment you hit rewrite, so the prompt knows what you actually sell, to whom, in your voice.

  4. That model's own guidance

    The rewrite follows the vendor's published prompting documentation for that exact target. Every change is traceable to it.

  5. The prompt you send

    With notes on what was optimized and why. Missing facts become marked placeholders, never invented ones.

The longer walkthrough, including what your vault can and cannot see

Image prompts are a different language.

Midjourney reads comma-phrases and flags. Nano Banana Pro wants a described scene. GPT Image wants composition spelled out. The same rough idea, rewritten for each.

Seedream, Nano Banana Pro, GPT Image, Midjourney, FLUX, Ideogram and more.

You type

product shot of my candle on a table, moody, make it look expensive

It sends

editorial product photograph of a hand-poured candle in a matte ceramic vessel, dark oak table, single warm side light, deep shadows, shallow depth of field --ar 4:5 --stylize 200 --no text, watermark

comma-phrase structure settings moved to flags --no instead of “don’t”

Example rewrites in the style each generator’s documentation recommends. Rewrites are AI-generated.

Prompts that know your business.

Save your projects and business context once, or point IterationFlow at your Obsidian vault. The details you’d otherwise retype every time ride along with every rewrite.

Saved once

Click one. Watch what it changes.

Every run of the Engine, automatically

write the spring email Write an email to Wattle & Ash subscribers announcing the new season’s single-origin is open for orders. Voice: plain and confident, no hype words. Audience: existing monthly bean subscribers.

Vault notes are read at the moment you rewrite, sent to the model only, and not stored by us. Any personal or business detail inside a prompt is handled the same way: passed through, never kept. How the context flow works

refactor this auth middleware and explain the tradeoffs

Model matchClaude Fable 5 would do this better. Multi-step code reasoning wants a thinking tier, not a fast one.

Max knows which model.

On the Max plan the Engine reads your prompt as you write it and points you at the model that will do that specific job best. One click to switch, before a single word is spent.

Making an image or a video? Paste your reference images straight into the Engine. The rewrite reads them, writes around what they already show, and tells you if a better reference would get a better result.

One added sentence took a model from 17.7% to 78.7% accurate.

Kojima et al., NeurIPS 2022. Prompt wording isn’t cosmetic. It is the difference between wrong and right.

up to 0 pts

Accuracy spread caused by formatting changes alone. Same model, same task.

Sclar et al., ICLR 2024, few-shot, LLaMA-2-13B
0%

Of AI users have abandoned a task and gone back to doing it by hand, because the output wasn’t usable.

Adobe, 2026 (n=1,008)
0%

Of AI users say they don’t know how to achieve the productivity gains expected of them.

Upwork Research Institute, 2024

Example rewrites in the style each model’s documentation recommends. Rewrites are AI-generated.

Simple plans.

Pick a plan and start rewriting in minutes. Your prompts are never stored. History lives only on your device.

Save 12% with yearly billing

Starter

A$15/month
Billed monthly
200,000 words a month
Choose Starter

Max

A$59/month
Billed monthly
1.5 million words a month
The Mac widget: the Engine one keystroke over any app.
Model Match: every prompt pointed at the model that will do it best.
Reference images: paste your refs and the rewrite reads them.
Choose Max

A word is a word we process: your prompt, any project context you attach, and the rewrite that comes back. A short rewrite runs about 500 words all in, so 500,000 words is roughly a thousand of them.

Enterprise

For organisations rolling IterationFlow out across teams. We scope the deployment with you, consolidate billing, and report on usage and optimizations by team. Priced on seats and volume.

Talk to us

Prices in AUD. Cancel any time; your plan runs to the end of the billing period.

Good questions.

Questions

Answer

Which models does it optimize for?

Ten text targets across Claude, GPT, Gemini, Grok, Llama and DeepSeek, with reasoning and standard variants handled separately. Image generators including Seedream, Nano Banana Pro, GPT Image, Midjourney, FLUX and Ideogram. Video models including Veo, Sora, Seedance, Kling, Wan, MiniMax, Runway and Luma.

Stop retyping. Start refining.