Plan your first content batch

Before you generate anything: the four things to fix on paper

Four decisions determine whether a batch hangs together, and none of them cost credits. Skip them and you will produce thirty technically fine images that do not read as one person.

The first is the niche — the recognisable territory the account occupies, specific enough that someone could describe it in a sentence. The second is the persona: who this character is, what they care about, how they speak. The third is a standout trait, one memorable, repeatable thing that makes the character identifiable at a glance in a crowded feed. The fourth is the look: a consistent palette, setting vocabulary and framing style that carries across every post so the feed reads as a whole rather than a folder.

Write these down before you open the generator. They become the constraints your prompts inherit, and constraints are what make a batch coherent. Generating without them is how you end up with a character who is somehow at a beach, in a studio, in a nightclub and on a mountain in the same fortnight, with no thread connecting any of it.

You already have ten images — check before you generate

If you built your character from scratch, the confirm step did more than train a model. It generated seven lifestyle photographs from your three reference shots, and those seven plus the three anchors — face, front, back — are sitting in your character's library right now. That is a ten-image head start you paid for as part of the 38-credit confirm.

They are not filler. The seven cover a deliberately varied range — different settings, different lighting, different framings — because that variety is what a training set needs. That same variety is what makes several of them directly postable, and at minimum they are an accurate preview of how your character reads in ordinary scenes rather than posed studio shots.

So the first action in planning a batch is to go and look at what exists. Some of the ten will be usable as-is, some will be reference material only. Either way you are not starting from zero, and knowing which of the ten you would actually post tells you a great deal about which directions to generate in next.

The target: thirty images and a handful of short videos

Aim for thirty still images and a small number of short videos before you open an account. Thirty is roughly a month at a daily cadence, or six to eight weeks at a more relaxed one, and it is enough to survive the period where you are learning what works without the pressure of producing under a deadline.

The videos matter disproportionately. Short video is what most platforms currently distribute hardest, and it is the format that reaches people who do not follow you yet, while stills are largely served to people who already do. Three or four short clips in the opening batch is a reasonable floor. You do not need thirty of them — video is more expensive to produce and slower to iterate on, and the sensible pattern is to learn what lands from your stills and then produce video against what you learn.

Do not aim for perfection across all thirty. Aim for coverage. A batch of thirty images where twenty-two are good and eight are mediocre is far more useful than fifteen carefully polished ones, because the twenty-two buy you the time to work out what your audience responds to — and that answer will change what you polish next.

How to split thirty images so the feed reads as a life

A feed that is thirty variations of the same shot reads as a catalogue. A feed that varies across a few clear registers reads as a person. The split below is a working starting point, not a law.

Roughly a third should be close-range: portraits, half-body, the shots where the face carries the image. These are what make the character recognisable and they are what people scroll past fastest, so they need to be the strongest.

Roughly a third should be full-scene: the character somewhere specific, doing something ordinary. Café, street, kitchen, park. These are what create the impression of a life rather than a photoshoot, and they are where your niche and standout trait should be visible.

The remaining third splits between two smaller groups. Some should be self-shot in feel — the casual, imperfect register that reads as a real phone photograph rather than a produced image; that unpolished quality is doing work, not failing to. And a few should be deliberate outliers: a different setting, a different time of day, something that breaks the pattern. Outliers are how you find out what your audience actually wants, and a batch with none of them teaches you nothing.

Across all four groups, hold the character's look constant and vary the world around them. That is the whole trick. Same person, different days.

What thirty images costs, and how to produce them

Once your character has finished training, generating with it is inexpensive. A standard-resolution image is 1 credit; the higher-resolution option is 2. So thirty images is 30 credits at standard resolution — less than the confirm step of the build itself.

The generator produces up to five images per run. That makes a run the natural unit of work: one prompt, one setting, five variations, of which you keep the two or three that landed. Thirty keepers therefore means somewhere between eight and twelve runs, and you should budget credits for the ones you discard rather than assuming a perfect hit rate. Forty to fifty credits is a realistic figure for thirty images you are actually happy with.

Work in themed runs rather than generating one image at a time. Set up a scene — a café in morning light, say — and produce five in that scene at once. You will get a coherent set that can be posted across several days, and you will spend far less time re-describing context. Batching by scene is also how you keep the palette consistent across the feed without thinking about it.

Generate at standard resolution while you are exploring, and reserve the higher-resolution option for images you have already decided to post. Paying double for a variation you are going to discard is the easiest credit leak to avoid.

Sequence it: the order that wastes the least

Produce in a sequence that lets each stage inform the next, rather than generating thirty images against assumptions you have not tested.

Start by reviewing the ten images the build already produced and marking which you would post. Then generate one themed run of five in your strongest scene and judge the character in a context you chose. If the results are wrong, the problem is upstream in your prompt vocabulary, and finding that out after 5 credits is much better than after 30.

From there, work through your shot types in themed runs until you have your thirty. Only then produce the short videos, using the stills you rated highest as the reference for what works — video is the most expensive thing in the batch, and producing it last means producing it against evidence.

Store the finished batch somewhere organised, tagged by shot type, before you post a single item. The point of the batch is that launch week is execution, not production. If you are still generating while you are trying to learn a platform's rhythms, you are doing two hard things at once and will do both badly.

Building your first batch

Roughly forty to fifty credits and an afternoon, assuming your character has finished training.

1. Fix the four decisions on paper — Niche, persona, standout trait, and a consistent look. No credits. Everything after this inherits them.
2. Audit what the build already gave you — Open your character's library. The three reference shots and the seven lifestyle photographs generated at confirm are there. Mark which you would genuinely post — that is your starting count.
3. Run one test scene — Generate a single run of five images in your strongest scene at standard resolution. 5 credits. If the character does not read right, fix the prompt vocabulary now rather than after thirty images.
4. Produce in themed runs — Work through your shot types — close-range, full-scene, self-shot register, a few outliers — five at a time in one setting per run. Keep two or three per run. Budget eight to twelve runs.
5. Upscale only the keepers — Regenerate at the higher resolution only for images you have decided to post. 2 credits each instead of 1. Never pay double for a variation you might discard.
6. Add three or four short videos last — Use your highest-rated stills as the reference for what works. Video is the most expensive item in the batch, so produce it against evidence rather than assumption.
7. Organise before you launch — Tag the finished batch by shot type and sequence it into a rough posting order. Launch week should be execution only.

The batch, in numbers

• 10 images you already have if you built from scratch — 3 reference shots plus 7 generated at confirm
• 30 still images as the launch target — roughly a month of daily posting
• 3 to 4 short videos, produced last, against what the stills taught you
• Roughly a third close-range, a third full-scene, the rest self-shot register and deliberate outliers
• 1 credit per standard image, 2 for higher resolution, up to 5 images per run
• 40 to 50 credits realistic for 30 images you are happy with, including discards

Frequently asked questions

How many images do I actually need before launching an account?

Thirty stills plus three or four short videos is a sound target, and it is less daunting than it sounds because you already have ten. If you built your character from scratch, the confirm step generated seven lifestyle photographs on top of your three reference shots, and they are sitting in the character's library. Thirty gives you roughly a month at a daily cadence, or six to eight weeks at a gentler one, which is the runway you need to learn what your audience responds to without simultaneously producing under deadline. At 1 credit per standard-resolution image and up to five per run, the generation itself is cheap — budget forty to fifty credits to account for the variations you discard. Coverage beats polish here. Twenty-two good images and eight mediocre ones is worth more than fifteen perfect ones, because the extras buy you time to learn.

Does the character builder produce any content I can post?

Yes, and most people miss it. When you press the final confirm on a from-scratch build, seven lifestyle photographs are generated from your three reference shots and used, together with those anchors, as the training set. Those ten images stay in your character's library. The seven are deliberately varied across settings, framings and lighting, because a training set needs range — and that same range means several of them are directly postable rather than merely technical. So the first step in planning a batch is not generating, it is looking. Go through the ten, mark the ones you would genuinely post, and start your count from there. Even the ones you would not post are useful: they show you how the character reads in ordinary scenes, which tells you where to aim the runs you are about to pay for.

How do I stop thirty images from looking like thirty different people?

Two habits. First, fix your look on paper before generating — palette, setting vocabulary, framing style — and let every prompt inherit it. Consistency in a feed comes from the world around the character being coherent, not just the face; a trained character will hold its face regardless, but a feed that jumps between a beach, a nightclub and a mountain in one week reads as chaos even with a perfectly consistent face. Second, generate in themed runs rather than one image at a time. Set up a scene, produce five in it, keep the two or three that landed, then move to the next scene. This gives you naturally coherent clusters you can spread across several posting days, and it keeps the palette consistent without you having to think about it. Vary the world, hold the person constant.

Should I produce videos in the first batch or wait?

Produce a small number, and produce them last. Short video is what platforms currently distribute hardest and it is the format that reaches people who do not already follow you, so a launch batch with no video is leaving your main reach lever unused. But video is the most expensive item in the batch and the slowest to iterate on, which makes it the wrong thing to guess at. The efficient order is to produce your thirty stills first, note which ones read strongest, and then generate three or four short clips built on that evidence. You will produce fewer wasted videos and better ones. After launch the ratio shifts — once you can see what your audience responds to, video deserves a larger share of your credits, because by then you are producing against data rather than assumption.

Related reading

• Defining your AI influencer's personality
• Where to post AI influencer content
• Running an AI influencer on Instagram