The moment Midjourney and Flux made a usable first still cheap enough that clients stopped treating generation as a scarce deliverable, the unpaid work shifted downstream — into the revision rounds that start after that first approval, when “just one more regen” still costs you CapCut rebuilds, re-exports across ratios, and evenings that never appeared on the rate card.

Faster gens did not raise your AI freelance rates. They moved the margin risk into the batch.

A decade ago, a first creative pass was expensive enough that clients paused before asking for another round. Render time, stock, photography, or a full motion pass made “try again” feel like a real request. Midjourney and Flux flipped that feeling. A fresh still arrives in minutes. Prompt adherence is good enough that the first frame often looks “close.” Close is dangerous for freelance AI pricing: it invites approval of the direction, then endless nudges on the same job.

The revision queue did not get cheaper. It got longer. After the client likes the first still, they treat every regen as free because the tool felt free. You still batch variants, rename exports, rebuild the CapCut timeline when the hero face drifts, and reopen Claude or ChatGPT only to rewrite a brief that was never locked. The invoice still looks like a flat project fee. The calendar looks like three clients and one exhausted evening.

That is the core ICP pain for a solo creative, motion freelancer, or small studio experimenting with AI: capacity is limited, retainers and project fees expect volume, and cheaper generations quietly inflate unpaid revision rounds after the first approval.

Flat fees made sense when deliverables were countable and revisions were scarce. Six stills. Two Reels. Two rounds of notes. Done meant done.

“Just one more regen” culture breaks that contract without rewriting it. The client approved the look on Tuesday. On Wednesday they want her hair darker. On Thursday the product angle changes. On Friday they ask for the same set in Stories, feed, and a 9:16 cut — each hop regenerating because nothing was locked as a kit. Each ask feels small. Together they eat the fee.

You cannot win by matching their speed with more Midjourney or Flux runs. Speed is what created the culture. Every unpaid regen trains the client that AI freelance rates buy infinite trial-and-error, not a scoped batch. Retainer clients are especially quick to learn this: monthly retainers start looking like open tabs, not capped revision rounds.

The agitation is not that clients are unreasonable. It is that your rate card still prices the first render while their behavior prices the infinite queue.

The first approved still is rarely where the hours go. They go after approval.

Batching: you generate five more angles so they can “pick.” Two land. Three are almost-right cousins. You rerun. Prompt adherence is light when the character was never locked — eyes shift, wardrobe drifts, lighting resets — so each batch is half salvage, half restart.

Re-exports: feed square, Stories vertical, carousel crop, maybe a thumbnail cousin. Each ratio is not a crop if the hero was generated once for one canvas. You regenerate or stretch, then rename, then drop into folders the client will never see but you will rebuild next week.

CapCut rebuilds: the Reel that used Monday’s still now needs Thursday’s face. Timeline markers move. Captions drift. Transitions that matched the old grade fight the new one. A “quick regen” becomes a half-hour edit tax, then another when they ask for a quieter hook.

None of that shows as a line item if you sold “AI content pack — flat.” The hours sit between approval and delivery — the gap cheap gens made easy to reopen forever.

Some freelancers try to signal modernity by listing Midjourney, Flux, CapCut, and Claude on the rate card. Seat names do not protect margin.

Clients do not buy your Midjourney plan. They buy outcomes: recognisable ads, a motion cut that matches last month’s hero, a pack that survives format hops. Naming tools invites the wrong conversation — “you already have Flux, so another variant is nothing” — instead of the right one: how many revision rounds, which formats, what locked identity, what “done” means.

FLUX.2 and Midjourney can both produce strong first stills; operating guidance and steps for faceless still batches is a separate production topic (https://havincy.com/blog/flux-ai-youtube-faceless-stills). For AI freelance rates, the seat is not the product. The product is a scoped batch against a locked kit. Listing seats without scope language tells the client the expensive part is generation. Generation is the cheap part. Continuity and revision discipline are the expensive parts — and they are invisible on a tool list.

Change what the fee covers.

Price a revision batch: a fixed number of rounds after first approval, with regenerations counted inside those rounds, not as infinite free nudges. Price multi-format packs as scoped deliverables — stills plus Stories plus one short cut — instead of “and then we’ll adapt.” Price the locked kit: persistent characters and universes so the face, wardrobe, and world do not restart from a blank Midjourney or Flux thread every Monday. That lock is what cuts unpaid regen churn; it is not a soft-sell feature list, it is how you stop selling the same cast twice.

Director-led produce light helps here: one pass that treats the approved still as source of truth for the formats in scope, not a vibe prompt reopened for every ratio. If the retainer renews on continuity, publish and measure only close what renews — which variants held, which face stays in the kit.

AI content freelancer margin recovers when the invoice names the batch and the lock, not the first pretty frame. Freelance AI pricing that still leads with “includes Midjourney generations” will keep losing to “just one more.”

Write scope the way AI clients actually behave.

Rounds: “Two revision rounds after first approval. Additional rounds billed at X.” Define a round as a consolidated note set, not a Slack drip of single regenerations. If they want drip, they buy drip.

Formats: list the canvases in the fee — e.g. six feed stills, three Stories frames, one CapCut short — and treat extra ratios as change orders. Multi-format packs only protect margin when the pack is named.

What “done” means: approval of the locked character or universe for the campaign, plus delivery of the named formats inside the round budget. “Done” is not “until it feels right in every possible crop.” Prompt exploration can sit in a paid discovery pass; production sits against the lock.

Claude or ChatGPT can help you draft that language into the brief. They do not replace the rate card. Put the same words in the proposal and the kickoff note so “just one more regen” has a documented home: inside a round, or outside the fee.

Blank-thread restarts are how unpaid revisions multiply: every job reopens Midjourney or Flux without a saved face, wardrobe, or world, so each client note becomes a full regen instead of a controlled variant. A client kit — locked characters and universes, multi-format outputs from that lock, Director-led produce against it — is what stops the Monday restart tax without turning this essay into another skills inventory. The longer walkthrough of that kit for freelancers lives here: https://havincy.com/blog/ai-tools-for-freelancers-client-kit. For AI freelance rates, the point is simpler: if identity is not locked, every revision round is priced like first generation even when you charge for it once.

Usually no — not as the primary unit.

Per-generation billing sounds precise and becomes a fight: which Midjourney or Flux run counts, what about near-duplicates, who paid for the failed prompt adherence pass. Clients also learn to ask for “just one more” as a micro-purchase war. Prefer packaging generations inside revision rounds and named batches. Cap exploration by time or a fixed variant count. Sell production as locked kit plus formats plus rounds. Scope the evening work — batching, re-exports, CapCut rebuilds — instead of metering every seed.

Agencies feel revision-loop pain across brands and headcount: continuity breaks between Midjourney, CapCut, Drive, and Ads Manager, and seats do not raise capacity when every note re-exports the brand. That agency problem is mapped here without cloning the playbook: https://havincy.com/blog/ai-tools-for-agencies-revision-loop. Solo freelancers and small studios feel a sharper version of the same economics. You have no bench. One unpaid regen queue after first approval displaces another client’s Thursday. AI freelance rates fail for you when the fee still assumes scarce revisions while clients assume infinite cheap gens. Agencies can hide some bleed across a team. You feel it the weekend you meant to take off. The fix rhyme is shared — lock identity, scope rounds, price the batch — on a rate card one person can defend.