PTN—22

Draft mode

A cheap, clearly-labeled draft pass for exploration; the expensive polished pass is an explicit upgrade.

Most generations are thrown away. Early in a task, people are exploring — testing a direction, checking a structure, seeing whether the idea survives contact with reality — and spending full-quality compute on output that exists to be discarded is waste dressed up as diligence. Draft mode makes the exploration tier official: a fast, cheap pass that is honest about being rough, with the polished pass held back as a deliberate upgrade for the survivors.

The label is the load-bearing element. A draft that looks like a final is a trap — it gets pasted into the deck, sent to the client, shipped — so draft output must be visibly, persistently marked as a draft, in a way that survives skimming. The economics should be stated at the fork, not discovered afterward: “Draft: seconds. Final: about a minute, uses a full credit” is a decision a person can actually make; two unlabeled buttons is a guess.

Refinement must be an explicit act. The moment a system auto-upgrades a draft it liked, the person has lost control of both the spend and the timing — and the entire point of the tier boundary evaporates. The same logic guards the exit: a draft should resist being sent or shipped by accident, with the affordance to finalize sitting exactly where the person would otherwise have exported rough work. Friction at the boundary between tiers is not bureaucracy; it is the pattern.

Done well, draft mode changes behavior upstream of any single generation. When exploration costs almost nothing, people try more directions, discard more freely, and commit polish to fewer, better candidates — the workflow every creative discipline already runs on sketches and thumbnails, finally priced correctly in the interface.

Anatomy

Fig. 01 — Draft modeLive specimen

Membership renewal email

Draft — rough passDrafted · 2 s · 1 credit

Your membership renews next month. Renew now to keep priority scheduling and the two seasonal tune-ups. Price is the same as last year.

Polished pass verifies figures against the membership records — about 40 s and 12 credits, versus 2 s and 1 credit for this draft.

Drafts can’t be sent — refine to final first.

A draft pass visibly marked as rough, with the cost of upgrading stated and refinement as an explicit action.

1Draft marking. Draft output is persistently labeled as a draft — it cannot be mistaken for a final at a glance.
2Stated cost difference. The latency and cost gap between draft and final is declared at the fork, not discovered after.
3Explicit refine action. Upgrading to the polished pass is a deliberate click, never an automatic escalation.
4Accident-proof exit. A draft resists being sent or shipped by mistake; finalizing is the gate out.

When to use

Generation is expensive — in latency, credits, or compute — and most outputs are exploratory.
The task has a natural sketch-then-commit rhythm: directions first, polish on the survivor.
People hesitate to iterate because every attempt bills like a final.
Quality tiers genuinely differ, and the difference is worth a decision rather than a default.

Design considerations

01Mark drafts persistently — in the output itself, not just the chrome around it — so the label survives a copy.
02State both tiers’ costs in the person’s currency: seconds, credits, message allowance.
03Carry the draft’s prompt, settings, and any edits into the refine pass; upgrading must not restart the work.
04Keep refinement scoped to the chosen draft — polish the survivor, not the whole spread.
05Put a confirmation on send/ship/export paths while output is still in draft state.
06Default new exploration to draft where the economics support it, with final as the opt-in.

Pitfalls

✕Drafts indistinguishable from finals — rough output escapes into decks, emails, and production because nothing marked it.
✕Auto-refinement: the system upgrades on its own schedule and the person discovers the spend on the invoice.
✕A refine pass that regenerates from scratch, discarding the draft the person had already shaped and approved.
✕Draft quality so poor it cannot predict the final, making the cheap tier a misleading preview rather than a sketch.

In the wild

Midjourney: Draft Mode renders at a fraction of the cost and wait of standard generation, with an explicit enhance step to re-render a chosen draft at full quality.
Runway: Turbo model tiers offer fast, cheaper video generation for exploring a shot before committing credits to the higher-fidelity pass.
Ideogram: Rendering speed options trade quality against cost per image, letting people iterate on prompts in the fast tier and finish in the slow one.