Tools
AI writing
By Ofrai Editorial (AI Model Curators)
Ofrai writing is multi-model chat: pick Claude for careful prose, GPT for a generalist pass, DeepSeek when you want cheaper drafts.
Chat and video use credits from your Ofrai balance.
When to switch
Outline on DeepSeek Flash, rewrite on Claude, keep GPT if the team already speaks that prompting dialect. All of that is Text Studio, not three websites. Route by the part of the job being done, because each stage has a different quality bar. Outlines and structure need breadth and cheap tokens, so they belong on Flash. A rewrite that must sound like a named person needs a model with a steadier long-form voice, which is the argument for Claude.
Factual work — checking a claim, tightening a paragraph, comparing two versions of the same sentence — goes to whichever model is fastest at reading and answering, because nobody wants the most expensive engine doing the easy half. The practical habit is a stop rule: run the cheap model until it stops improving, then escalate once for the rewrite. Escalating at the first weak draft is the most common way people end up paying premium rates for outline work.
Pair with media
After the copy is stable, open Image or Video Studio with the same credits for the accompanying still or clip. Sequence matters more than it looks. Write the headline first, because the headline decides the composition of the hero image; write the body last, because paragraphs rarely change what the image should show. For a landing page, one strong still above the fold beats a strip of five small ones, and the still costs one generation while the strip costs five.
For social, keep the copy platform-specific: a caption is not a headline, and an image that fits a feed post usually fails as a story because the text sits outside the safe area. The credits come from the same balance as the writing session, so a long drafting afternoon has already reduced the budget you have for the visuals it was meant to support.
When to switch to Image or Video Studio
After the prose locks, open Image or Video Studio from this page and pair the copy with a still or a short clip using the same credits. No second login. Switch when the words are approved, not when they are merely written, because a visual built on a headline that later changes is wasted work on both sides. Two rules keep the pairing cheap. Generate the still at the ratio the placement actually uses, so nothing has to be re-cropped.
And produce the clip from an approved still rather than from text, because a clip inherits its frame from an image you have already reviewed. If the visual turns out to need the argument to be different — often a screenshot needs the feature named differently than the prose does — go back and change the copy instead of designing around the mismatch.
When the prose locks, switch studios
When you are happy with the copy, hop into Image or Video Studio from this page. The thread and credit balance stay put, so stills and clips go on the same Ofrai account.
Rewriting without losing your own sentences
The most common frustration in a long writing session is the machine editing away the lines you liked. Three instructions prevent it. First, mark the keepers: paste back the two or three sentences that worked and say they are fixed. Second, ask for a diff, not a replacement — "return only the paragraphs you changed, with the first four words of the original beside them" makes the change reviewable and stops the model quietly rewriting the opening you never asked it to touch.
Third, scope the edit: "tighten this section by fifteen percent without removing any example or claim" is a brief, while "make it better" is a licence to replace everything. When a draft is mostly your own words, editing sentence by sentence beats regenerating sections. When it is mostly machine text, regenerating beats polishing, because you will spend an hour removing filler that should never have been written.
Voice profiles you can paste into any model
Every writing model has a default house voice, and it is rarely yours. Override it with a short block you paste at the top of the thread: audience, reading level, sentence length, what to avoid, and two sentences you wrote yourself as a reference. "Audience: procurement managers at a mid-market software company. Reading level: a senior operator, not an academic. Average sentence under 20 words. No throat-clearing, no rule of three, no em-dash as a tic. Reference: these two paragraphs."
That block costs a fraction of a full draft and stops the rewrite loop before it starts. Keep one file per brand voice and paste the same file into every model, so switching from DeepSeek to Claude does not silently reset your tone.
Outline first, prose second, on long drafts
Long pieces fail when you ask for prose in one shot. Work in three passes. Pass one, ask for a bare outline with a one-line promise per section and an explicit word budget — that alone removes half the filler. Pass two, write each section separately with the outline line pasted as the instruction, which keeps sections from repeating each other. Pass three, paste the sections in order and ask for one pass focused only on transitions and length, not on new ideas.
This is also where the model tier matters: a cheap Flash model is fine for outline and section drafts, while the careful model is worth its cost for the transition pass and the final read. Doing it this way is faster than regenerating a 1,500-word draft four times.
Editing passes that tighten machine prose
Machine output has a signature: it is competent and slightly airless. Four passes fix most of it. Ask for cuts only — "remove 15 percent without losing meaning". Ask for specificity — replace every abstraction with a number, a name, or a date. Ask for varied sentence openings, because models default to participial clauses. Ask for one moment of roughness: a real objection, a concrete example, an aside a human would actually have. Run these in that order on the same text.
Skipping the specificity pass is why editing AI copy often feels like sanding; specificity is what makes it sound written rather than assembled.
Failure modes: openings, hedges, and filler
Watch for four tells. Hollow openings — "In today's fast-paced world" — always: delete the first sentence and see whether anything is lost. Hedge stacking — "may potentially suggest" — usually hides a claim you should either make or drop. Rule of three everywhere, because the model reaches for triads when it has nothing specific. And consensus padding, where a paragraph restates the previous one in longer words. The fix is mechanical: paste the paragraph back and ask "what is new in this paragraph relative to the one above?"
If the model cannot answer in one sentence, the paragraph should not exist. Repeating these prompts on every draft costs a few credits and most of the editing time.
Token cost and choosing which model to bill
Text models bill on tokens, so cost follows length, not clicks. That makes the cheap-versus-expensive decision easy to reason about: use the Flash tier for anything you will throw away — outlines, first drafts, bullet variants, summarising your own notes — and the careful tier only for work you will keep, such as a final pass on copy that ships.
A 900-token rewrite costs a fraction of what a full 8,000-token generation costs on the same model, which is why editing an existing draft is usually the cheapest way to raise quality. Long pasted context is the other cost driver: attach a brief, not a whole archive. Ofrai charges the same credits across text, image, and video, so a runaway chat session competes directly with your media budget.
A 45-minute long-form article workflow
Minutes 0–10: paste the brief and your voice profile, ask for an outline with word budgets, and cut the sections you would not write. Minutes 10–25: draft the two sections that carry the argument section by section, on the cheap tier, with the outline line as the instruction. Minutes 25–35: escalate to the careful model for one transition-and-tightening pass over the whole draft.
Minutes 35–45: generate the on-page assets — a hero still in Image Studio, a short clip if the article needs one — while the credits for the draft are still fresh. The draft, the passes, and the assets stay in one Ofrai thread and one library, so the next article starts from the voice profile instead of from zero.
When an AI writing tool is the wrong choice
Skip it for work where being wrong is expensive or personal: legal and medical copy, contracts, safety instructions, anything with a number a regulator will check, and news about real named people. Also skip it when the deliverable is a decision, not a document — strategy memos and hiring judgements need your judgement in the room, not a draft to react to. And skip it if you need guaranteed originality in a form the model will not give you, such as a personal essay written in your own experience.
Models are strong at structure, rewriting, summaries, and drafts. They are weak at facts they cannot check and at anything that requires having been there.
Facts, claims, and citations you must verify
Models state numbers confidently and invent them happily. Treat every statistic, date, price, quotation, and study name as an unverified draft until you have seen it in a primary source. Ask the model to mark which claims it is confident about and which it inferred, but treat that split as a hint, not a guarantee — confidence and accuracy are different things. Verify in this order: anything with a number attached, anything you would be embarrassed to retract, and anything legal.
Keep a note of where each verified fact came from next to the draft, so the next editor does not repeat the work. If a piece cannot survive verification, the numbers are usually the wrong story, not the writing.
Open studio
FAQ
- Is this ChatGPT?
- No. Ofrai lists those engines in one place, including GPT-class models when they are available.
- Can I compare Claude and GPT here?
- Yes — see the Claude vs GPT page, then switch inside Text Studio.
- How much does a generation cost?
- Each job lists Ofrai credits in the studio. Catalog credits are starting points; resolution and duration can change the final spend.
- How do I start on Ofrai?
- Open Image or Video Studio from this page, pick a recommended model, paste or adapt the starter prompt, and generate from your library.
- What is AI writing on Ofrai?
- Multi-model chat in Text Studio: Claude for careful prose, GPT for a generalist pass, DeepSeek for cheaper drafts. Switch inside one place with one credit balance.